research paper in marine science

Aims and scope

Marine Life Science & Technology (MLST), launched in 2019, publishes original research papers with new discoveries and theories across a broad range of life sciences and technologies, including basic biology, fisheries science and technology, medicinal bioresources, food science, biotechnology, ecology and environmental biology, especially associated with marine habitats. The journal puts emphasis on fostering synergistic interactions among the broad disciplines mentioned above, and aims to promote multidisciplinary approaches within the field of sciences. The journal is intended for biological scientists, aquaculture researchers, marine technologists, biological oceanographers and ecologists.

Specifically, but not exclusively, MLST covers the following topics:

• Basic Biology : all biological disciplines, including aquatic (both marine and freshwater) biodiversity, biochemistry, cell biology, computational biology, development, evolution, genetics and epigenetics, immunology, microbiology, omics and bioinformatics, phycology, physiology, reproductive biology, and zoology, etc .
• Fisheries Science & Technology : fisheries management and ecosystem modeling, stocking and enhancement technologies, fish physiology, reproduction and breeding, feeding and aquafeed development, immunology and disease control, stock assessment and environmental sustainability, etc .
• Marine Drugs & Bioproducts : including the discovery, structure elucidation, chemical synthesis, semi-synthesis, biosynthesis, structure-activity relationship and pharmacology of bioactive compounds from marine and terrestrial natural materials.
• Food Science & Biotechnology : including seafood nutrition, seafood processing and storage, food safety control and testing, food microbiology and biotechnology, functional foods and activity evaluation, marine enzyme engineering and biomaterials.
• Ecology & Environmental Biology : including biological oceanography, aquatic and terrestrial ecology, restoration ecology and technology, ecotoxicology, ecological security, environmental biology and health.
• Find a journal
• Publish with us
• Track your research

Khaled bin Sultan Living Oceans Foundation

Providing science-based solutions to protect and restore ocean health

Skip to content

Living Oceans Foundation

You are here:

• Publications

Scientific articles

Remotely sensed habitat diversity predicts species diversity on coral reefs

Anna C. Bakker, Arthur C.R. Gleason, Alexandra C. Dempsey, Helen E. Fox, Rebecca H. Green, Sam J. Purkis, "Remotely sensed habitat diversity predicts species diversity on coral reefs," Remote Sensing of Environment, Volume 302, 2024, 113990, ISSN 0034-4257, https://doi.org/10.1016/j.rse.2024.113990.

KSLOF's Chief Scientist, Dr. Sam Purkis, worked with his Ph.D. student, Dr. Anna Baker, to develop a new way to use satellites to analyze global reef biodiversity.

The 2022 Hunga-Tonga Mega-tsunami: Near-Field Simulation of a Once-in-a-Century Event

Sam J. Purkis, Steven N. Ward, Nathan M. Fitzpatrick, James B. Garvin, Dan Slayback, Shane J. Cronin, Monica Palaseanu-Lovejoy, and Alexandra Dempsey. "The 2022 Hunga-Tonga Mega-tsunami: Near-Field Simulation of a Once-in-a-Century Event." Science Advances.

KSLOF's Chief Scientist, Dr. Sam Purkis, used bathymetry data collected on the Foundation's Global Reef Expedition to model the size of the tsunami to hit Tonga in January of 2022. He and found that it was similar in size to the one caused by the eruption of Krakatoa in 1883.

Expediting the Search for Climate-Resilient Reef Corals in the Coral Triangle with Artificial Intelligence

Mayfield, Anderson B., Alexandra C. Dempsey, Chii-Shiarng Chen, and Chiahsin Lin. 2022. "Expediting the Search for Climate-Resilient Reef Corals in the Coral Triangle with Artificial Intelligence" Applied Sciences 12, no. 24: 12955. https://doi.org/10.3390/app122412955

Anderson Mayfield, a former KSLOF Fellow, used data collected on the Foundation's Global Reef Expedition to develop a machine-learning approach for identifying climate-resilient corals in the Solomon Islands.

Heat, human, hydrodynamic, and habitat drivers measured from space correlate with metrics of reef health across the South Pacific

Bakker, A.C., Gleason, A.C.R., Mantero, A. et al. Heat, human, hydrodynamic, and habitat drivers measured from space correlate with metrics of reef health across the South Pacific. Coral Reefs (2022). https://doi.org/10.1007/s00338-022-02325-9

This paper in Coral Reefs, utilized the Living Oceans Foundation’s Global Reef Expedition field dataset to build a model that can predict coral cover and other metrics of coral reef health using open-source satellite data.

A New Foraminiferal Bioindicator for Long-Term Heat Stress on Coral Reefs

Humphreys, A.F., Purkis, S.J., Wan, C. et al. A New Foraminiferal Bioindicator for Long-Term Heat Stress on Coral Reefs. J. Earth Sci. 33, 1451–1459 (2022). https://doi.org/10.1007/s12583-021-1543-7

KSLOF Chief Scientist, Dr. Sam Purkis, worked with our partners at the University of Miami to develop a way to use benthic foraminifera as bioindicators for reef health.

Tsunamigenic Potential of an Incipient Submarine Landslide in the Tiran Straits

Purkis, S. J., Ward, S. N., Shernisky, H., Chimienti, G., Sharifi, A., Marchese, F., et al. (2022). Tsunamigenic potential of an incipient submarine landslide in the Tiran Straits. Geophysical Research Letters, 49, e2021GL097493. https://doi.org/10.1029/2021GL097493

KSLOF Chief Scientist, Dr. Sam Purkis, released a study of the potential impacts of a tsunami in the Red Sea.

Environmentally-Driven Variation in the Physiology of a New Caledonian Reef Coral

Mayfield, A.B.; Dempsey, A.C. Environmentally-Driven Variation in the Physiology of a New Caledonian Reef Coral. Oceans 2022, 3, 15–29. https://doi.org/10.3390/oceans3010002

This publication by one of our former fellows and our Director of Science Management studies the physiology of 'pristine' corals in New Caledonia just before the onset of severe annual bleaching events, so that future generations might know how these corals functioned in their last bleaching-free year.

Go to page:

• Privacy Overview
• Strictly Necessary Cookies
• 3rd Party Cookies
• Cookie Policy

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.  You can view our complete Privacy Policy here .

Most of our cookies are used to improve website security and reduce spam. These cookies should be enabled at all times. They also enable us to save your preferences for cookie settings.

This website uses Google Analytics to collect anonymous information such as the number of visitors to the site, and the most popular pages. Keeping this cookie enabled helps us to improve our website.

Please enable Strictly Necessary Cookies first so that we can save your preferences!

More information about our Cookie Policy

• Search Menu
• Advance articles
• Editor's Choice
• Food for Thought
• Food for Thought: Luminaries
• Food for Thought: Rising Tides
• Quo Vadimus
• Stories from the Front Lines
• Symposium Issues
• Themed Sets
• Introductions
• Hidden Gems
• Author Guidelines
• Submission Site
• Open Access
• Why Publish with us?
• About ICES Journal of Marine Science
• About the International Council for the Exploration of the Sea
• Editorial Board
• Advertising and Corporate Services
• Self-Archiving Policy
• Dispatch Dates
• Terms and Conditions
• Journals on Oxford Academic
• Books on Oxford Academic

Article Contents

What is machine learning and why does marine ecology need it, a quick primer on machine learning, the setup of the database and its tags, machine learning to extract ecological information from observational data, machine learning to improve ecological understanding, discussion and perspectives, acknowledgement, conflict of interest, authors contributions, data availability, machine learning in marine ecology: an overview of techniques and applications.

• Article contents
• Figures & tables
• Supplementary Data

Peter Rubbens, Stephanie Brodie, Tristan Cordier, Diogo Destro Barcellos, Paul Devos, Jose A Fernandes-Salvador, Jennifer I Fincham, Alessandra Gomes, Nils Olav Handegard, Kerry Howell, Cédric Jamet, Kyrre Heldal Kartveit, Hassan Moustahfid, Clea Parcerisas, Dimitris Politikos, Raphaëlle Sauzède, Maria Sokolova, Laura Uusitalo, Laure Van den Bulcke, Aloysius T M van Helmond, Jordan T Watson, Heather Welch, Oscar Beltran-Perez, Samuel Chaffron, David S Greenberg, Bernhard Kühn, Rainer Kiko, Madiop Lo, Rubens M Lopes, Klas Ove Möller, William Michaels, Ahmet Pala, Jean-Baptiste Romagnan, Pia Schuchert, Vahid Seydi, Sebastian Villasante, Ketil Malde, Jean-Olivier Irisson, Machine learning in marine ecology: an overview of techniques and applications, ICES Journal of Marine Science , Volume 80, Issue 7, September 2023, Pages 1829–1853, https://doi.org/10.1093/icesjms/fsad100

• Permissions Icon Permissions

Machine learning covers a large set of algorithms that can be trained to identify patterns in data. Thanks to the increase in the amount of data and computing power available, it has become pervasive across scientific disciplines. We first highlight why machine learning is needed in marine ecology. Then we provide a quick primer on machine learning techniques and vocabulary. We built a database of ∼1000 publications that implement such techniques to analyse marine ecology data. For various data types (images, optical spectra, acoustics, omics, geolocations, biogeochemical profiles, and satellite imagery), we present a historical perspective on applications that proved influential, can serve as templates for new work, or represent the diversity of approaches. Then, we illustrate how machine learning can be used to better understand ecological systems, by combining various sources of marine data. Through this coverage of the literature, we demonstrate an increase in the proportion of marine ecology studies that use machine learning, the pervasiveness of images as a data source, the dominance of machine learning for classification-type problems, and a shift towards deep learning for all data types. This overview is meant to guide researchers who wish to apply machine learning methods to their marine datasets.

The term “machine learning” (ML) has become omnipresent in both the scientific literature and everyday news. Its first use dates back to the late 1950s: Regarding a game of checkers, Arthur Samuel, an electrical engineer at IBM, stated that “a computer can be programmed so that it will learn to play a better game of checkers than can be played by the person who wrote the program” by using so-called “machine-learning procedures” (Samuel, 1959 , p. 219). In its broadest definition, an ML system improves its performance by extracting information from data (Mitchell, 1997 ). In contrast to traditional computer programs, which encode a solution designed by the programmer, an ML system can learn to solve a task without being provided an explicit recipe. Instead, the task is learned by providing the system with examples, i.e. data. The ability to produce a solution to a problem that is not representable mechanistically can be extremely powerful, but it depends crucially on selecting an appropriate representation of the problem (an “objective” function) and on having adequate data from which to learn.

Although often used interchangeably in popular literature, ML is a subdomain of the larger field of artificial intelligence (AI), which encompasses knowledge representation, logic models, algorithms, and computational methods capable of intelligent behaviour ( Figure 1 ). Within ML, the subfield of deep learning (DL; LeCun et al ., 2015 ) has advanced rapidly over the last decade. DL systems use large neural networks ( Table 1 ) to extract relevant features from raw data and learn from them, instead of requiring explicit engineering of those features. These data are often complex (such as images or sounds) and big (thousands to millions of records). In this review, we cover ML, therefore, including DL.

Deep learning is a subdomain of machine learning, which on its own is a subdomain of artificial intelligence, as illustrated. Specific methods are mentioned in each subdomain.

Definitions of machine learning algorithms commonly used in marine ecology studies and cited in this review.

The success of ML is associated with the increase in computational power over the last 20 years (Mitchell, 1999 ), but also with the increasing volume of available data (Jordan and Mitchell, 2015 ), which led to the development of a broader diversity of algorithms, implemented in widely available software. Scientists from many disciplines outside of computer science are now actively applying ML methods, and marine sciences are no exception, as exemplified by a recent themed set in this journal (Beyan and Browman, 2020 ). Most examples in this themed set actually relate to ecological questions, within which a central focus is the detection and quantification of the abundance and distribution of living organisms. ML is promising in marine ecology for several reasons. (i) Modern instruments produce large volumes of data (Tanhua et al ., 2019 ; Guidi et al ., 2020 ) that require scaling up their processing; the flexibility and adaptability of ML methods make them a natural choice for such automation. (ii) This automation can also help to reduce the biases necessarily introduced by manual processing (e.g. Culverhouse et al ., 2003 ), hence improving reproducibility. (iii) Finally, ML is adept at handling high degrees of uncertainty (i.e. dealing with noise in the data) associated with unknown underlying mechanisms or with non-stationary processes; therefore, they often yield high predictive power (Baker et al ., 2018 ) and are increasingly used to gain an understanding of ecological processes (Lucas, 2020 ).

Within marine sciences, ML has been used more extensively in some subdomains. Specialized reviews have already covered some applications. For example, Liu and Weisberg (2011 ) reviewed the use of Self-Organizing Maps in Oceanography, Culverhouse et al . (2006 ), Benfield et al . (2007 ), and Irisson et al . (2022 ) reviewed ML techniques for the taxonomic classification of plankton images, Reichstein et al . (2019 ) gave an overview of DL for earth sciences—including oceanic applications, and Malde et al . (2020 ) provided a brief review on recent developments in DL and highlighted both opportunities and challenges for its adoption in marine sciences.

For researchers whose expertise is outside of computer sciences, ensuring a proper application of ML methods and keeping track of new developments is challenging. The aim of the present review is to serve as a resource for marine ecologists who want to apply ML to their own data. To that effect, the section “A quick primer on machine learning” serves as a starting point for non-practitioners and introduces relevant vocabulary. The section “The setup of the database and its tags” describes our survey of the literature and the resulting structured database, on which the rest of the review is built. From it, we identified that ML is used at two stages in ecological research: (i) to process the raw data collected and extract ecologically meaningful datasets from it and then (ii) to combine these ecology-ready datasets together, and with others, to improve our understanding of ecological systems. Therefore, the section “Machine learning to extract ecological information from observational data” describes applications where ML was used to generate ecological datasets from various raw data types: images and video, optical spectra of single cells, acoustics, omics, geolocation records, and ocean colour imagery and biogeochemical profiles. The section “Machine learning to improve ecological understanding” describes how ML can be used to gain knowledge on the relationships between species and their environment (section “Predicting species abundance and distribution”), among species (section “Capturing dynamic ecological relationships”), and between us, humans, and marine ecosystems (sections “Summarizing ecosystems through regionalization” and “Supporting human decisions on ecosystem management”). Finally, the section “Discussion and perspectives” concludes on the commonalities among ML applications, suggests what is currently limiting them in ecology, and gives a general outlook of the field.

In this section, we provide a short overview of the different tasks that ML can achieve, the overall process that ML studies go through in the context of marine ecology, and then present different ML algorithms and software tools that implement them. Interested readers are invited to consult classic texts to deepen their understanding, either in an introductory manner (James et al ., 2013 ) or in a more mathematically oriented one (Friedman et al ., 2001 ).

ML approaches are often divided between supervised and unsupervised. Supervised systems are given a set of input data points and their corresponding output (measurements or labels assigned by experts). The output is often called the target or response variable. In this case, an ML system learns the mapping from the input variables to the output variable (e.g. predict fish diversity from environmental variables; Smoliński and Radtke, 2017 ). Supervised systems can further be divided into classification, where the output is categorical and the task is to assign a class to input data (e.g. classify plankton taxa from images; Gorsky et al ., 2010 ), and regression, where the output is continuous or at least ordered (e.g. predict nutrient concentrations from hydrological variables; Sauzède et al ., 2017 ). A supervised task relevant to marine ecology is object detection: The ML system locates objects of interest in a form of regression, often of their bounding box (e.g. detect benthic organisms in images of the seafloor; Liu and Wang, 2021 ). Finally, sometimes, the target variable is only available for a subset of data points, a situation called semi-supervised learning.

Unsupervised systems are given input data only and search for patterns without the availability of a target variable. For instance, unsupervised methods can aim to cluster data points together based on a definition of similarity (e.g. define distinct bioregions based on community compositions; Sonnewald et al ., 2020 ), to define simpler representations for the data while retaining salient properties, also known as dimensionality reduction (e.g. represent correlations between environmental variables through the first two dimensions of a principal component analysis; Zhao and Costello, 2019 ), or to construct a model for the distribution of the data (e.g. produce a smooth map of the density of active fishing vessels from point records; Kroodsma et al ., 2018 ).

Additional steps can be performed before or within an ML pipeline. An important part of many ML systems is the preprocessing (e.g. feature normalization or smoothing) of input variables, in order to make them as relevant as possible. Feature extraction derives new informative variables from initial, raw ones (e.g. automated extraction of measurements from an image; Hu and Davis, 2005 ). Feature selection eliminates less relevant variables, either to improve performance or to gain explainability thanks to a simpler system (e.g. removal of correlated variables; Thomas et al ., 2018 ). Finally, the covariance structure in the input variables can be used to impute missing values, which are common in field-collected data, or detect outliers, i.e. values that go beyond the expected range of covariance (e.g. a dissolved oxygen concentration too high given the temperature of the water).

The general process for tackling an ML task is shown in  Figure 2 , and the successive steps are described in its caption. Of course, depending on the approach and study case, some steps will be modified. For example, target variables are not available in the case of unsupervised learning (step 2). In DL, feature extraction is included in the model (step 4). In many situations, cross-validation is used in lieu of a dedicated validation set (step 6): The training set is split into subsets, the model is trained on all subsets but one, and validated on this remaining one; this process is repeated until each subset has been held out once. Comparisons with an external dataset (step 8), although important, are rarely performed due to the lack of such independent data. Finally, many ML models are never deployed (step 9), but serve to describe and understand a particular dataset.

The general process of (supervised) machine learning. After being collected (1), data need to be labelled (2), which means associating the inputs with a number or a name as output ( l  = 1, 2, or 3 in the example). The data are then split into training, validation, and test datasets (3) while taking its structure into account (e.g. ensure that all labels are represented in each dataset). Each input in the training set can be summarized into features (4). The (transformed) training set is used to train the model (5), by minimizing a loss function (L) that computes the value of one or several performance metrics (M). The validation set undergoes the same transformation as the training set, if any, and is then used to evaluate the predictive performance of the model, ideally with the same metric(s) (6). Several versions of the model can be trained with different hyperparameters (i.e. settings, noted h * ) of the machine learning system, and the one with the best performance on the validation set is retained. At this point, the model is frozen and its final performance is assessed on the test set (7). If external information, different from the original data, is available, it should be used to ensure that model predictions are reasonable, in addition to achieving a given performance (8). Finally, the model is ready to be deployed and used with newly collected data (9).

Diverse ML algorithms have been developed to solve a large variety of tasks. In  Table 1 , we provide a brief description of those commonly used in marine ecology publications.

Finally, several open-source software libraries implement many ML methods under a consistent interface. Thus, once one understands the general process (as highlighted in  Figure 2 ), exploring various methods is relatively easy and progress can be quick. The better known libraries of relevance for marine ecology are scikit-learn ( https://scikit-learn.org/ ; Pedregosa et al ., 2011 ) and, more recently, TensorFlow ( https://www.tensorflow.org ; Abadi et al ., 2016 ) and PyTorch ( https://pytorch.org ; Paszke et al ., 2019 ) in Python, the tidymodels collection of packages in R ( https://www.tidymodels.org/ ; Kuhn and Wickham, 2020 ), Flux ( https://fluxml.ai/ ) in Julia, or Weka in Java ( https://www.cs.waikato.ac.nz/ml/weka/ ).

As a basis for this paper, we built a database of literature references covering the application of ML methods to marine data (supplemented by a few additional works, outside of this scope, but providing context and cited in this review). In its broadest definition, ML covers a wide array of methods and data types. Because many methods have been applied in marine ecology, it is extremely challenging to make an exhaustive inventory. Therefore, the goal of this database is instead to showcase the diversity of ML applications to marine data. To do so, multiple keyword-based searches in various scholar databases were performed by the authors, including the keywords “machine learning”, “marine”, “ecology”, and variations thereof. The results were complemented with the personal libraries of the authors, who span a range of specialties. This (already large) nucleus of papers was further grown using the references cited within them, starting from the most recent and going backwards in time. This procedure was iterative (the references of the newly added papers being also examined) and the search was stopped after several rounds of such tentative additions did not yield any new reference.

After assembling this large body of potentially relevant literature, the authors screened the suitability of each paper for its inclusion in the database according to the following criteria: (i) the paper is peer-reviewed, (ii) its “Methods” section describes the ML approach used, and (iii) it applies it to a marine dataset. Because some classical statistical techniques can be perceived as ML, we further reduced the scope to studies that follow the general process of  Figure 2 (i.e. include a validation and/or a test dataset). Then, papers were organized through tags, defining the type of data they analyse (“data:*” tag), ML tasks achieved (“task:*”), the algorithms used (“method:*”), and other useful characteristics (e.g. availability of code and/or data; “meta:*”). The content of the resulting database, organized according to data type, is summarized in  Figure 3 .

Treemap representation of the papers in the database that can be categorized according to the type of data they use. The area of each rectangle is proportional to the number of papers (written in brackets). The broad data types are bold and coloured with a given hue. Sub-types, when they exist, are in variations of the same hue.

This selection process still yielded over 1000 papers, which cannot all be described in this review. To decide which ones to cite, we considered the following additional criteria: the paper (i) has been widely adopted by the research community (e.g. is cited very often, defines a method widely applied), (ii) is easily reproducible because its methodology is well-described and/or code and data are publicly available, or (iii) is representative of a body of work not covered by criteria (i) or (ii).

While this database is not exhaustive, the methodical approach described above should avoid overt biases and large omissions. We therefore consider it representative of the diversity of approaches and of the relative volume of research in various domains. More importantly, we hope its use will become continuously maintained and updated by its users. To do so, users can browse the library online ( https://www.zotero.org/groups/2325748/wgmlearn/library ) and, if they wish to contribute, register to the WGMLEARN Zotero group ( https://www.zotero.org/groups/2325748/wgmlearn/ ), indicating what their contribution would be. The library in its state at the time of submission is available as Supplementary Material (S1) .

A first set of extensive and successful applications of ML is the processing of raw inputs (images, sounds, sequences, etc.) into ecologically meaningful data, often in the form of tables with samples (locations, times, etc.) as rows and variables (taxa densities, biogeochemical quantities, etc.) as columns.

Quantifying marine objects from images and video

Methods to segment and classify objects of interest from images or video are not sensitive to whether the object is a fish, a bird, or a piece of plastic debris. Yet, the processing of this dominant ( Figure 3 ) type of data has a long history that is often siloed within specific communities, sometimes with reason. For example, object segmentation is very different for benthic objects lying over a complex background than for pelagic ones, imaged over a rather uniform background. Therefore, the literature is presented separately for benthos, marine macrolitter, nekton, and plankton. The commonalities among the methods used for these data, and others, are highlighted in the “Discussion and perspectives” section.

Underwater imaging of the benthic environment has grown considerably in the last few decades. We reviewed over 100 papers that used ML to process such data and all were published after 2000 (the earliest is Soriano et al ., 2001 ). Almost half of these focused on habitat mapping (e.g. Porskamp et al ., 2018 ) and coral reefs and their inhabitants (e.g. Villon et al ., 2018 ), followed by studies focussing on the detection of benthic invertebrates (e.g. Kiranyaz et al ., 2010 , 2011 ). The most used algorithms included support vector machines (SVMs), random forest (RF), convolutional neural networks (CNNs), k-nearest neighbours (kNN), and classification and regression trees (CARTs). Those were used mostly for image/pixel classification (in more than half of the studies) on their own or with other algorithms, as previously pointed out by Lopez-Vazquez et al . (2020 ). More recently, object-based classification has replaced pixel-based classification (Zhang et al ., 2013 ), especially using CNNs, which reached much higher performance (Gómez-Ríos et al ., 2019 ; Piechaud et al ., 2019 ).

Growth in this field has naturally been accompanied by an increase in the number of images of benthic fauna and habitats. Though ML offers promise towards unlocking the catalogue of unused benthic images, many challenges remain. There are growing concerns regarding the identification of available data for training, the pre-training of deep nets, and the handling of class imbalance in training datasets. For example, of the millions of images acquired each year on coral reef surveys, just 1–2% are labelled (Beijbom et al ., 2012 ). Lumini et al . (2020 ) compared several CNN architectures and found that combinations of several models (i.e. ensembles) were the most successful for image classification of coral (and plankton) datasets. Fincham et al . (2020 ), who classified images from across multiple benthic habitats, found an imbalance in their training data due to the frequency of habitat occurrence, which was countered by using data augmentation to artificially expand the training by flipping, scaling, and rotating images.

The challenge in accessing high-quality training datasets is now being addressed through developments such as standardized reference catalogues (Althaus et al ., 2015 ; Fisher et al ., 2016 ; Howell et al ., 2019 ), wide adoption of specialized annotation software such as BIIGLE 2 (Langenkämper et al ., 2017 ), SQUIDLE+ (Williams and Friedman, 2018 ), and CoralNet (Beijbom et al ., 2015 ), and annotated image databases e.g. FathomNet (Boulais et al ., 2020 ). In addition, the development of user-friendly software such as VIAME and Superannotate is making ML more accessible to benthic ecologists. However, for researchers to best apply these tools, much remains to be learned regarding model performance under different conditions (e.g. depending on the number of classes used), on training dataset size, on the use of single models versus ensembles of models, etc. (Durden et al ., 2021 ).

Macrolitter

Each year, tonnes of human-created waste litters the sea surface, seafloor, and shorelines and poses a major threat to oceanic ecosystems and coastal communities (NOAA, 2014 ). Extensive surveys and research are conducted worldwide to assess litter distributions and concentrations in coastal areas and the open sea, to identify litter accumulation zones through numerical models, and to design management actions to promote litter removal and recycling (NOAA, 2016 ; Madricardo et al ., 2020 ). To quantify marine litter, video and photography-based monitoring is increasingly adopted and deployed on bottom trawl or nets, autonomous underwater vehicles, remotely operated vehicles, unmanned aerial systems, and drones. However, litter identification is mainly done by humans, which is time-consuming, costly, and often very subjective, creating the need for automatic approaches (Canals et al ., 2020 ).

Region-based convolutional neural networks (R-CNNs), designed for object detection, have been increasingly applied to automatically detect and classify beached (Watanabe et al ., 2019 ), floating (Lieshout et al ., 2020 ), and seafloor (Politikos et al ., 2021 ) macrolitter items. Additionally, traditional CNN classifiers have been used to categorize litter types from segmented images (Garcia-Garin et al ., 2021 ). Such studies have generally shown that the classification and detection performance of neural networks is high for floating litter (>80%) but often lower for underwater and seafloor litter, which can be attributed to the challenges of underwater imagery (various camera angles, zoom levels, light shadings, litter buried in the seabed). Several authors have used open and experimental datasets for their analysis, focusing mainly on the predictive performance of the algorithms. The applicability of ML for marine macrolitter research has been recently reviewed in more detail (Politikos et al ., 2023 ).

Ultimately, DL has the potential to support monitoring of marine litter by providing automatic, rapid, and scalable solutions. Nevertheless, a collection of images and video recordings from real-world environments and more effective algorithms are needed to support litter assessment goals set by stakeholders (Politikos et al ., 2023 ). Finally, new imaging technologies such as infrared detection (Inada et al ., 2001 ) or Raman imaging (Gallager, 2019 ), which can identify plastics at least in a laboratory setting, could be implemented and integrated with ML techniques for improved results.

Monitoring of nekton informs decision-making for biodiversity conservation and sustainable fisheries management. Imaging surveys constitute a non-invasive complement to conventional monitoring. However, it yields large datasets and ML has come into play to automate and speed up the data processing. Nekton monitoring from images is challenging due to the diversity of tasks that need to be solved (e.g. species classification but also morphometric estimations) and the very different conditions in which images are collected (e.g. both underwater and on ships).

In early fish imaging studies, classic ML methods were used with data obtained in controlled, experimental setups. For example, in Storbeck and Daan (2001 ), the image acquisition system consisted of a camera and a laser, which allowed obtaining images but also information on fish volume. They classified six species of fish with 95% accuracy using a shallow artificial neural network (ANN) based on fish contour features. In Zion et al . (2007 ), three edible fish species were sorted using a minimum Mahalanobis distance classifier that combined geometric features and object contours as inputs to yield an accuracy of >96%.

In situ monitoring of nekton is largely focused on fish as well, but those studies present additional challenges due to the wide variations in observation conditions. Datasets are typically collected with underwater cameras (Fisher et al ., 2016 ), but larger organisms, such as marine mammals, are also monitored via satellite images. Here also, before the development of CNNs, global features were used together with background modelling to detect and track objects under water. For example, Spampinato et al . (2010 ) developed a Gaussian mixture model (GMM) and moving average algorithm followed by an adapting mean-shift algorithm to detect and track fish in in situ videos, with an 85% success rate. Hu et al . (2012 ) reached >97% accuracy in the classification of fish images based on texture and colour features, using two kinds of SVMs. Another approach to handle the change in appearance of objects underwater is to consider the information from a sequence of frames in a video rather than from only one frame, as done in Shafait et al . (2016 ) where, for ten species, accuracy ranged from 71 to 100%. Finally, artificial alterations in images (i.e. data augmentation) are a common way to improve performance and generalization in CNNs. Allken et al . (2019 ) trained an Inception 3 architecture on 5000 data-augmented images per species to reach 94% accuracy on a test set, while the baseline model, trained on the 70 original images, reached an accuracy between 50 and 71%. Bogucki et al . (2019 ) used a combination of three CNNs to detect and identify North Atlantic right whales in aerial and satellite images. The first CNNs located the whales in satellite images, the second detected key points on the whales’ heads in aerial survey images, and the third identified the whales.

Images of nektonic organisms are also collected outside of the water, by camera systems deployed on fishing vessels (known as electronic monitoring), which have replaced some on-board fishery observers. Deep models have become valuable tools to process the videos collected (Helmond et al ., 2020 ). Specifically, Mask R-CNN was provided pixel-level masks and bounding boxes around organisms to automatically monitor catches on-board fishing vessels (Tseng and Kuo, 2020 ). Such masks can be used to automatically measure the length of the detected objects. In Garcia et al . (2020 ), segmentation performance was assessed with the intersection over union (IoU) metric computed between the predicted and ground truth masks. In this study, 1605 images were used to train a Mask R-CNN model, and an average IoU of 0.89 was obtained on 200 independent test images. Notably, in more challenging images where fishes overlapped, the IoU was lower, as expected.

Because planktonic organisms are often micrometric to millimetric, high magnification is needed to image them, which implies a short depth of field and can lead to many out-of-focus objects. In addition, in situ , images are dominated by morphologically diverse detrital particles that are similar in size to living organisms. Finally, the organisms themselves are also incredibly diverse. Therefore, the automatic classification of such images is a difficult and interesting ML problem, and remains a major bottleneck for their exploitation.

The first attempts at machine-based classification of plankton images derived various features from the images: statistical moments (which capture size, average lightness, etc.), Fourier transforms of the contour of the object, texture patterns (Tang et al ., 1998 ), and, later, grey-level co-occurrence matrices (Hu and Davis, 2005 ). Those features were input into a classifier, often an ANN (Tang et al ., 1998 ), an SVM (Hu and Davis, 2005 ), or a combination of both to classify images into a limited (mostly fewer than ten) number of taxa.

These approaches matured and the next decade saw the rise of their application for numerous ecological studies. The most influential papers of this period are associated with popular instruments and software. For instance, Grosjean et al . ( 2004 ) and Gorsky et al . (2010 ), while presenting the ZooScan, highlighted that (i) the performance of different classifiers is largely similar and therefore mostly determined by the original features, (ii) this performance decreases strongly when the number of taxa to classify increases, and (iii) with 8 taxa, predictive power saturates beyond 300 example images per taxon in the training set. Sosik and Olson (2007 ) presented the Imaging FlowCytoBot and described in detail the reasoning and process to derive features particularly relevant for phytoplankton, from the original images. Despite the large number of papers, applications of those techniques at broad spatial and temporal scales are still rare (but see Irigoien et al ., 2009 ).

The next evolution in this research was the increasing use of CNNs, particularly since 2015, owing to a plankton image classification competition run on Kaggle.com (Robinson et al ., 2017 ;  Figure 4 ). However, the thoroughness of papers using this technique is inconsistent and many are published in conference proceedings that are difficult to access. By contrast, Ellen et al . (2019 ) provide an extensive overview of the setup of a CNN from scratch, including the choice of its parameters, the inclusion of classic image features and other external information into the CNN’s classifier, and compare the CNN’s performance with the more classical approaches described above.

Classifiers used for plankton image recognition through time. The plot displays the proportions rather than absolute numbers. The time bins on the x -axis are not regular. The plot highlights the quick adoption of support vector machines and their current decline, the rise and fall of random forests, the increase of convolutional neural networks (particularly since 2015), and their current dominance.

CNNs will likely be increasingly relied upon in the future. Their implementation within dedicated plankton imaging software such as EcoTaxa (Picheral et al ., 2017 ) or the IFCB dashboard will facilitate their routine use by a wide community of ecologists ( https://ifcb-data.whoi.edu/dashboard ). The separation of their feature-extraction part from their classification part seems like a promising avenue for transfer learning (i.e. using a model initially trained on one, often general, dataset to quickly “fine-tune” it on a another dataset, here a plankton one; Orenstein and Beijbom, 2017 ), unsupervised classification (Schroeder et al ., 2020 ), and active learning procedures, whereby only few images representative of the diversity of the dataset are shown to the user (Bochinski et al ., 2019 ).

An important problem in plankton image datasets, like in many other biological ones, is class imbalance (a few classes dominate the samples). Among several solutions, generating synthetic images in the rare classes using a generative adversarial network (GAN) was recently tested (Li et al ., 2021 ). Alternatively, quantification approaches, which do not aim to perfectly classify each individual object but rather to directly derive concentration estimates for each class, deal intrinsically with the distribution among classes (Gonzalez et al ., 2019 ).

Identifying microorganisms from single-cell spectra

Flow cytometry has been used since the 1990s to study marine microbial communities. In flow cytometry, scatter and fluorescent properties of individual particles are measured at very high rates (i.e. hundreds to thousands of particles per second). Although most researchers manually analyse the resulting “cytograms”, automated methods have become available for the analysis of such microbial flow cytometry data (Rubbens and Props, 2021 ).

Since the 1990s and early 2000s, artificial neural networks (ANNs) have been developed to identify up to 72 lab-grown phytoplankton species using flow cytometry (Boddy et al ., 2001 ). Supervised single-cell classifiers were then successfully applied for the identification of heterotrophic bacteria as well, by combining flow cytometry with a nucleic acid stain in most cases. Besides ANNs, support vector machines (SVMs), linear discriminant analysis (LDA), and random forests (RFs) have been successfully applied in this setup (Rajwa et al ., 2008 ; Rubbens et al ., 2017 ). These lab-based studies have demonstrated the usefulness of the information captured by flow cytometry for bacterial and phytoplankton identification. However, it is difficult to transfer this knowledge directly to samples taken from the field. As the identity of species present is often unknown, labels are not available to train supervised models. When analysing field samples, unsupervised clustering approaches are therefore used to group together cells that have similar optical properties. Examples include Gaussian mixture models (GMMs), graph-based clustering, and self-organizing maps (SOMs) (Hyrkas et al ., 2016 ; Sgier et al ., 2016 ; Bowman et al ., 2017 ).

In some cases, cell populations do not form distinct patches that can be isolated by clustering: when the complexity of microbial communities is high (i.e. many taxa) or the resolution is limited (e.g. due to the instrumental setup or when studying heterotrophic organisms). Cytometric fingerprinting approaches do not try to identify cell populations; instead, they focus on modelling the multivariate distribution of the cytometric data, by defining informative regions in this distribution and recording cell counts or densities in those regions. Often, binning approaches are employed, although more advanced strategies have become available as well, e.g. by overclustering the data using a GMM (Rubbens et al ., 2021 ) or by an automated deleting, merging, and shrinking of Gaussian mixtures (Bruckmann et al ., 2022 ).

A few hybrid approaches have been proposed for freshwater samples, in which information from laboratory cultures is used to analyse natural samples. RF classification was used to differentiate noise from signal using lab-grown cultures and then used to remove the noise in natural samples (Thomas et al ., 2018 ). Learned representations of lab-grown cultures can also be used as proxies to describe the dynamics of a microbial community in a natural sample (Özel Duygan et al ., 2020 ).

Raman spectroscopy is an alternative, information-rich, single-cell technology for the identification of marine microorganisms. Spectra typically contain many more variables than traditional flow cytometry data; therefore, the use of convolutional neural networks (CNNs) should be beneficial to summarize this information and get to single-particle identification. When a CNN was trained on Raman spectroscopy data, it resulted in high classification accuracy for 13 marine microorganisms (∼95%) but similar to that of SVM and LDA, probably due to a low sample size (Liu et al ., 2020 ).

Describing ecosystems with acoustics

Light attenuates faster in water than in air, limiting cameras to observing a small volume (albeit at high resolution). Sound propagates over long distances and is used to monitor the ocean interior. Sound also samples larger volumes of water than towed nets and can be used in areas that are otherwise difficult to reach, such as deep water and rough bathymetry. Both active sensors, which emit sound and measure the returned echoes, either from organisms in the water column or from the seabed, and passive sensors, which just “listen”, are commonly used in marine science. The following text is organized along those categories.

Active acoustics for target classification

Active acoustics are widely used in fisheries and aquaculture to evaluate the spatial and temporal distributions of organisms, measure their size distribution, and calculate population structure, as well as characterize the behaviours of species. In all cases, the analysis starts with the identification of the returned echo, also called target classification (Korneliussen, 2018 ). This process frequently involves manually checking, cleaning, processing, and scrutinizing the echogram features. Target objects are then delineated and ascribed to species using “expert” knowledge gained from biological samples. This heavy dependence on manual operations makes the process time-consuming and vulnerable to bias; scalable and reproducible methods, such as ML-based approaches, are therefore needed.

Early attempts to automate target classification typically used deterministic features computed from the data, using details in individual echo pulses (Rose and Leggett, 1988 ) and/or school-based features like shape or energy, as well as auxiliary information like location or depth; this information was passed to a range of classifiers. Artificial neural networks (ANNs) were used early on (Cabreira et al ., 2009 ;  Figure 5 ). Random forests (RFs) were used with school-based features and auxiliary information, for fish identification (Fallon et al ., 2016 ). Peña ( 2018 ) recently reviewed clustering techniques for acoustic data and concluded that expectation-maximization (EM) clustering is the only technique that properly separates acoustic signatures (and noise), after a supervised initialization.

Evolution of the methods used for target classification from active acoustics data, through time. The labels give the total number of references per row or column of the plot. The colour is proportional to the number of references published in the time period of the column and using the method of the row. A single reference can appear in several rows if it uses several methods.

Wideband or multi-frequency echosounders added the frequency dimension to the data, which allowed for improved discriminatory power. Using the frequency response usually involved averaging over certain ping- or range-bins and comparing the scatter distributions to the properties of known aggregations. Using the full broadband echo spectrum, an RF classifier was successful in classifying individual fishes (Gugele et al ., 2021 ).

More recently, convolutional neural networks (CNNs) were used to classify the entire echogram and identify the primary species on patches of echo (Hirama et al ., 2017 ;  Figure 5 ). Shang and Li ( 2018 ) used simulated data to compare different classifiers using different features and CNNs reached the best performance. Regions of interest, identified on real data, were more accurately identified by CNNs with various architectures (ResNet, DenseNet, Inception) than by a support vector machine (SVM) classifier working on traditional manual features (Rezvanifar et al ., 2019 ). CNNs have also been used for pixel-level predictions (i.e. segmentation) on raw acoustic data, using a U-net architecture (Brautaset et al ., 2020 ) or Mask-Regional CNN (Marques et al ., 2021 ), trained on manually labelled data. Such supervised methods require large amounts of training data, while recently developed semi-supervised methods allowed only ∼10% of the training data to be labelled (Choi et al ., 2021 ).

Active acoustics for seabed and sediment mapping

Active acoustics are also used to map seabed topography and sediment cover, which condition the type of benthic biological community that can develop. Various methods reach high spatial resolutions and accuracy, such as single-beam echosounders, sidescan sonars, and reflection sismographs, but multi-beam echosounders (MBES) are the most cost-effective for mapping large areas (Anderson et al ., 2008 ; Brown et al ., 2011 ). Bathymetry and backscatter data (and their derivatives) are interpreted in order to characterize the type of seabed substrate. For a thorough description of conventional sea bottom classification systems, see the extensive work of Hamilton (2001 ).

One major challenge for seabed mapping is that the manual interpretation of seabed features from acoustic data is very time-consuming and highly subjective. This explains the increased interest for automated approaches, including inversion algorithms, image-processing techniques, and, mostly, ML (Brown et al ., 2011 ; Stephens and Diesing, 2014 ).

In the 1990s, the early ML approaches were ANN, e.g. Stewart et al . (1994 ), who successfully classified three different seafloor types based on sidescan sonar data. Dartnell and Gardner (2004 ) used hierarchical decision trees (DTs) trained on four types of images (backscatter intensity and three variance images). Using 60 ground truth sediment samples, they predicted seafloor types in Santa Monica Bay with an accuracy of 72%, which was better than other automated classification methods at the time.

Since then, a variety of ML methods have been scrutinized through comparative studies (Ierodiaconou et al ., 2011 ; Stephens and Diesing, 2014 ; Shao et al ., 2021 ). The classification algorithms were very diverse, covering tree-based methods (DT, random forest—RF, Quick Unbiased and Efficient Statistical Tree—QUEST, and Classification Rule with Unbiased Interaction Selection and Estimation—CRUISE, etc.), SVMs, maximum-likelihood classifiers (MLCs), and ANNs. In many cases, ML-based approaches were not significantly different from one another but were vastly superior to the usual manual interpretation procedures.

Recently, Cui et al . (2021 ) demonstrated how a deep belief network (DBN) based on fuzzy ranking feature optimization can be used to map sediment distribution over large areas. Fuzzy ranking is a technique used to identify the feature combination, derived from the MBES data, that is most appropriate for the DBN to correctly classify the seabed sediment type. The accuracy of the DBN proved higher than that of five other supervised classification models (DT, RF, SVM, MLC, and ANN).

Passive acoustics monitoring

Passive acoustic recordings are a reliable and cost-effective method to monitor habitat use, distribution, density, and behaviour of species over space and time. They can be obtained from boats, autonomous devices (either fixed or moving ones), cabled stations, and animal tags, making them usable in a variety of situations (Kowarski and Moors‐Murphy, 2021 ). However, because of their relative ease of use, hydrophones quickly generate large datasets that require automation to extract information from them (Gibb et al ., 2019 ).

The most common approach to process acoustic data is to detect and classify specific acoustic events in a supervised manner. Sound source classification studies have primarily focused on shipping (Zaugg et al ., 2010 ) and mammals’ vocalizations (66 out of the 101 references we recorded; Bittle and Duncan, 2013 ). In the latter, detection and classification algorithms have been used to identify species (Bermant et al ., 2019 ), specific calls (Bergler et al ., 2019 ), or even dialects and individuals (Brown et al ., 2010 ). ML can also be used to localize the position of, or estimate the range to, a certain source without the need to model the sound propagation (Niu et al ., 2017 ), outperforming conventional matched field processing methods. Another application is to relate properties of the source with characteristics of the sound, through regression; these properties included the size of male sperm whales (Beslin et al ., 2018 ) or fish abundance (Rowell et al ., 2017 ). In addition, ML can be used for acoustic source separation, a problem known as the cocktail party problem (Bermant, 2021 ). Finally, approaches to characterize entire habitats from their soundscape have also been explored (Lin et al ., 2019 ).

A common approach is to extract human-engineered features from the sound and use them as input for an ML algorithm. These features can be derived from the time, frequency, or cepstral domain (transformation of the data to highlight periodic signals), or based on the full image of the spectrogram, a visual representation of sound intensity per frequency as a function of time (Sharma et al ., 2020 ). The algorithms used for classification include SVMs (Jarvis et al ., 2008 ), RFs (Malfante et al ., 2018 ), Gaussian mixture models (GMMs; Roch et al ., 2011 ), and k-means (Weilgart and Whitehead, 1997 ), among others. More focus has been put on identifying which features are relevant for the classification and characterization of sound events than on which classifier performs best. Often these features or other rule-based signal processing techniques are also used to first segment the data and then ML is used to classify the detected segments.

Advances in image and speech-recognition algorithms have been applied to underwater sound, reducing the amount of preprocessing and improving performance and generalizations (Schröter et al ., 2019 ). In DL approaches, sound is often converted into a spectrogram, which is considered as an image and input into a convolutional neural network (CNN) for classification, regression, or feature extraction and clustering (Bermant et al ., 2019 ; Thomas et al ., 2020 ). However, recently some models have been developed that are applied directly on the waveform (Roch et al ., 2021 ).

In the marine context, sounds of interest can be very sparsely occurring and datasets can comprise long periods of time. This leads to highly imbalanced datasets. This imbalance is usually solved by first detecting and then classifying the detected sounds, where the detection step is a rule-based signal-processing algorithm and the classification step is a DL approach (Stowell, 2022 ). However, the biggest limitation for the application of ML to passive acoustic recordings is the lack of knowledge regarding which sounds are produced by which species, because visual surveys to associate sound with images of the species are often impossible. This leads to a lack of data annotation and limits the usage of supervised ML approaches. To compensate for the lack of ground-truth data, unsupervised clustering algorithms are being developed to acquire general information about the ecology of certain habitats (Ozanich et al ., 2021 ).

Profiling biological communities with environmental genomics

The study of nucleic acids obtained from an environmental sample is coined as environmental genomics (or meta-omics). In marine ecology studies, the genetic information usually comes from a community of organisms rather than from a single specimen, which is our focus here. Metabarcoding (amplification by polymerase chain reaction and sequencing of a taxonomically informative gene) allows documenting biological communities in terms of species presence and proportions. Metagenomics (shotgun sequencing of a complex mixture of genomic DNA) provides information of random sections of genomes, allowing us to gain insight into both taxonomy and functions. Metatranscriptomics (shotgun sequencing of isolated RNA transcripts) provides similar information for genes active at the time of sampling.

ML approaches have long been used for genomics data analysis. This includes both translating raw signals into nucleotides using base-calling algorithms (Wick et al ., 2019 ) and sequence data analysis. For instance, hidden Markov models have been extensively used for functional annotations, multiple sequence alignments (Yoon, 2009 ), and more recently for viral signatures detections in metagenomic datasets (Ponsero and Hurwitz, 2019 ). However, few studies have applied ML to strictly marine meta-omics data. We therefore provide a general overview of the analysis of metabarcoding data and highlight some ML applications to marine data.

Metabarcoding datasets are usually processed by well-established bioinformatics software, e.g. QIIME 2 (Bolyen et al ., 2019 ), which translates raw sequences into statistically exploitable species-to-sites count matrices. Sequences are often grouped into operational taxonomic units (OTUs) or amplicon sequences variants (ASVs) based on their similarity. These sequence units then serve as a proxy for species/strains to document biodiversity changes. Current algorithms to cluster sequences into OTUs or ASVs are VSEARCH (Rognes et al ., 2016 ), which relies on an arbitrary similarity cutoff to delineate OTUs (e.g. 97%), SWARM (Mahé et al ., 2015 ), which aggregates neighbouring sequences to abundant, supposedly genuine, seed sequences, or DADA2 (Callahan et al ., 2016 ), which uses base calling values to separate spurious from genuine sequences. These two latter methods find more “natural” boundaries of OTUs and, as such, can be considered as unsupervised approaches. Some OTUs are then assigned a taxonomic name based on similarities with known sequences from curated databases (e.g. PR2, Guillou et al ., 2013 ; SILVA, Quast et al ., 2013 ). To this end, several ML-based methods have been developed, including naive Bayes (NB) classifiers (the RDP classifier, Wang et al ., 2007 ) and classification trees using k-mers distributions across sequences (Murali et al ., 2018 ). More recent work successfully applied convolutional neural networks (CNNs) to process and taxonomically annotate raw metabarcoding data faster, without relying on operational OTUs or ASVs (Flück et al ., 2022 ).

Resulting OTU-to-site count matrices are then amenable to biodiversity analysis using compositionality-aware multivariate statistics (Quinn et al ., 2019 ). For example, ML allows routine monitoring of the impact of industries on marine biodiversity. Based on metabarcoding datasets labelled with ecological states obtained by conventional methods, random forest (RF) models can be trained to assess the ecological status of new samples, based on their metabarcoding profiles alone. This is faster and more cost-effective than conventional morpho-taxonomy approaches, enabling scaling up the spatio-temporal scales of biomonitoring programs (Cordier et al ., 2018 ; Frühe et al ., 2021 ).

Network ecology research has been developed on interactions between macro-organisms (e.g. plant-pollinator interaction networks), but many interactions remain difficult to observe and validate. This is especially true within microbial communities, for which statistical frameworks have been developed to detect co-occurrence patterns and include them into more holistic ecological studies. ML techniques can be used to predict species interactions (Vacher et al ., 2016 ; Bohan et al ., 2017 ) and can outperform the identification of trait-matching combinations compared to generalized linear models (Pichler et al ., 2020 ). Microbial networks can be inferred from genomics data (Faust and Raes, 2012 ; Lima-Mendez et al ., 2015 ) as a means to predict putative biotic interactions, which opens new avenues for understanding the links between marine microbial communities and the large-scale functioning of marine ecosystems (Guidi et al ., 2016 ; Chaffron et al ., 2021 ). Finally, ML is expected to contribute to improve our capacity to analyse massive meta-datasets composed of numerous collated cross-study genomics data, by controlling for covariates (Wirbel et al ., 2021 ).

Quantifying and mapping fishing pressure from geolocation data

Fishing and shipping activities are putting important pressure on marine ecosystems. They are often tracked using vessel monitoring systems (VMSs) or the automatic identification system (AIS), which transmits vessel locations at regular intervals (Thoya et al ., 2021 ). VMSs are required by fisheries management agencies for many commercial fishing vessels and the data are often confidential. AIS is designed for maritime safety, for any type of vessel, and the data are more broadly accessible. These data are often extensively processed using ML to identify vessel and gear types (Russo et al ., 2011 ; Marzuki et al ., 2018 ; Taconet et al ., 2019 ).

Many studies have classified fishing vs. non-fishing behaviours using artificial neural networks (ANNs; Bertrand et al ., 2008 ; Russo et al ., 2014 ) and random forests (RFs; Ducharme-Barth and Ahrens, 2017 ; Behivoke et al ., 2021 ). To do so, the movement characteristics of vessels across space, time, and habitats are often studied and summarized before being provided to the ML classifier. Kroodsma et al . (2018 ) trained convolutional neural networks (CNNs) with AIS data to identify fishing vs. non-fishing behaviours and fishing gear types, producing the first map of the global footprint of fisheries (Taconet et al ., 2019 ).

The outputs of these models have been used not only to assess fishing pressure but also in ecological studies to estimate noise impacts (Allen et al ., 2018 ), assess marine spatial planning or monitor conservation areas (Robards et al ., 2016 ; White et al ., 2020 ), identify species distribution (Le Guyader et al ., 2016 ), minimize mammal strike risk (Fournier et al ., 2018 ), and mitigate bycatch (Richards et al ., 2021 ).

To integrate fishing activity with the rest of the ecosystem, ML efforts on fishery geolocation data have used an expanded suite of predictor variables. For example, several studies used boosted regression trees (BRTs) to relate fishing locations with environmental information (e.g. sea surface temperature) and then predict dynamic maps of fishing activity from environmental data (Soykan et al ., 2014 ; Crespo et al ., 2018 ). Other studies added bio-economic considerations into fisher location-choice frameworks, with ANNs (Dreyfus-Leon and Kleiber, 2001 ; Russo et al ., 2019 ). By characterizing fishing behaviours using these broader features (e.g. environment, bio-economics), ML approaches provide a valuable foundation for operational, dynamic, ocean management tools that support ecosystem-based fishery management in near real-time (Hazen et al ., 2018 ).

Deriving biogeochemical variables from satellite images and floats profiles

Historically, most in situ measurements used for the characterization of ocean biogeochemical processes were acquired using ships, resulting in critical undersampling at a global scale. Advances in remote sensing (by ocean colour satellites) and in situ robots now allow sampling marine bio-optical variables at unprecedented spatio-temporal resolution (Claustre et al ., 2020 ).

Yuan et al . (2020 ) provide a review of applications of DL to environmental remote sensing for estimating atmospheric, land, and oceanic physical, chemical, optical, and biogeochemical variables. One section is dedicated to the use of ML for remotely sensed ocean colour parameters retrieval, mainly focussed on the estimation of the chlorophyll-a concentration. However, ML has also been applied to remote-sensing data to derive fields of inherent optical properties of the seawater (Ioannou et al ., 2011 , 2013 ), p CO 2 (Landschützer et al ., 2015 ), primary production (Mattei et al ., 2018 ), phytoplankton community composition (Stock and Subramaniam, 2020 ), particulate organic carbon (Liu et al ., 2021 ), dissolved inorganic carbon (Roshan and DeVries, 2017 ), and nitrogen fixation rate (Tang et al ., 2019 ), as well as perform atmospheric correction (Jamet et al ., 2005 ; Brajard et al ., 2012 ) ( Figure 6 ).

Machine learning methods used with satellite imagery data. Artifical neural networks (in blue shades), and, in particular, multi-layer perceptrons, dominate the literature that was reviewed.

One remarkable example of using ML for ocean science is the synergy between satellite observations and in situ profiles, in particular from the Argo programs (>100000 currently). Sauzède et al . (2016 ) used a multi-layer perceptron to extend surface bio-optical properties to depth. This produces four-dimensional (i.e. longitude, latitude, depth, and time) fields of biogeochemical variables at global or regional scales, which fill in situ observational gaps. Such continuous fields are particularly valuable for the initialization and validation of biogeochemical models. They are now reaching operational status since four-dimensional fields of chlorophyll-a concentration and particulate organic carbon generated by these methods have recently been made publicly available on the European online portal Copernicus Marine Environment Monitoring Service.

Finally, ML methods are also used to estimate the more scarcely measured biogeochemical variables from the more commonly measured physical ones. For example, an ANN was trained to predict nutrient concentrations and carbonate system variables from over 250000 profiles of pressure, temperature, salinity, and oxygen concentration (Bittig et al ., 2018 ). The predictor variables can be measured with very high accuracy by autonomous floats and now ANN-based methods can spatially and temporally populate the fields of nutrients and carbon variables, which were previously loosely resolved. MLPs have also been used to predict the phytoplankton community composition from profiles of fluorescence of chlorophyll-a (Sauzède et al ., 2015a ), making it possible to gather and homogenize tens of thousands of fluorescence profiles available from historical databases, which could not be integrated in global analyses before (Sauzède et al ., 2015b ).

Once ecology-ready tables of data have been extracted from raw sources (see section “Machine learning to extract information from observational data”), they can be analysed to gain a better understanding of socio-ecological marine systems (this section). Such studies traditionally use statistics, often multivariate, and modelling to capture relationships between observed variables; this task is also amenable to ML. In this section, we highlight how ML techniques are used to relate species to their environment and, in particular, predict species distributions, detect dynamic interactions involving several species, and, finally, inform ecosystem management by partitioning the environment in easier-to-understand units through regionalization and fueling monitoring and decision-support tools.

This field is even more difficult to map through literature searches than the more technical studies presented in the previous section. Some searches with relevant keywords yielded >10000 results, while others with minor differences yielded only hundreds. Therefore, in this section, even more than in the previous one, we really focus on presenting papers that showcase different approaches.

Predicting species abundance and distribution

The ability for ML approaches to capture complex and non-linear relationships, as well as their ability to work with missing and heterogeneous data, has driven their popularity for the analysis of species–environment relationships.

When data are sparse or heterogeneous, often also leading to high uncertainty, Bayesian ML methods have proven useful. Fernandes et al . ( 2010 ) predicted fish recruitment using a naive Bayes (NB) classifier relying on spawning stock biomass, climate, and weather data. Fernandes et al . (2013 ) used multi-dimensional Bayesian networks for a similar task and found that predicting three species simultaneously doubled the chance of being correct, compared to three single-species models. Lehikoinen et al . (2019 ) used tree-augmented NB models to evaluate the influence of various environmental factors, all heterogeneous in type and in spatio-temporal resolution, on coastal fish abundance. They note that some environmental factors are not relevant to predict average abundances, but are important for extreme ones.

Tree-based ensemble models such as random forests (RFs) and boosted regression trees (BRTs) have also proven useful with ecological data thanks to their versatility and ease of use. Knudby et al . ( 2010 ) found tree-based methods superior to linear models in predicting species richness, biomass, and diversity in coral reefs based on habitat variables. Suikkanen et al . (2021 ) used RF regression to analyse the relationships of zoo- and phytoplankton (particularly cyanobacteria) in multidecadal (but relatively sparse) monitoring data to find whether relationships found in experiments could also be seen in field data.

Species distribution models (SDMs) are frequently applied to perform spatially explicit analyses of ecological data. They quantify the relationship between species occurrence or abundance and their environment and can be then used to predict their potential geographical distribution (Guisan and Thuiller, 2005 ; Elith and Leathwick, 2009 ). A significant body of literature compared the performance of ML-based SDMs with multivariate linear regression or climate envelope methods, generally finding that ML methods yield better predictive performance but are prone to overfitting (e.g. Derville et al ., 2018 ).

The most widely applied ML method for SDMs is Maxent, with over 6000 published papers, which showcases the power and broad applicability of ML for ecological inference (Phillips and Dudík, 2008 ; Elith et al ., 2011 ). Maxent works with records of a species present at given points in space and iteratively maximizes the probability of presence at these points, predicted from functions of environmental variables at the same points (Phillips et al ., 2006 ). But, many other ML approaches are also used in species distribution modelling, such as decision trees (DTs; Hunt et al ., 2020 ), BRTs (Elith et al ., 2008 ; Cimino et al ., 2020 ), RFs (Reiss et al ., 2011 ), support vector machines (SVMs; Knudby, 2010 ; Vestbo et al ., 2018 ), and artificial (Benkendorf, 2020 ) and convolutional neural networks (CNNs; Deneu et al ., 2021 ). These models have been applied to resolve a diverse range of ecological and conservation issues, including understanding species ecology (Brodie et al ., 2018 ), responses to current and future environmental change (Hindell et al ., 2020 ), threat overlap (Welch et al ., 2018 ), and the design and evaluation of spatial management scenarios (Stock et al ., 2020 ; Smith et al ., 2021 ). Across all applications, communicating the uncertainty of SDMs to stakeholders is critical. In general, estimating uncertainty within ML-based SDMs is difficult, and most solutions underestimate model uncertainty (Beale and Lennon, 2012 ; Watling et al ., 2015 ; Brodie et al ., 2020 ). However, new approaches, such as Bayesian additive regression trees, are emerging and improving our estimation of uncertainty (Carlson, 2020 ).

Capturing dynamic ecological relationships

As climate variability and long-term change drive non-stationarity in ecosystems, more research is needed to see how ML approaches can improve our ability to predict and forecast potentially changing species relationships with their environment and other species. Latent (hidden) variable modelling provides one way to detect an underlying systemic change, or to approximate an ecosystem component that is not represented in the dataset. Trifonova et al . ( 2015 ) modelled the North Sea ecosystem using dynamic Bayesian networks with hidden variables (DBN-HVs), and concluded that a hidden variable in the model managed to learn the zooplankton biomass variations in all modelled areas. Trifonova et al . (2017 ) used this model to predict ecosystem responses under different scenarios. Uusitalo et al . (2018 ) and Maldonado et al . (2019 ) created a DBN-HV model for the central Baltic Sea food web and found that the hidden variables replicated the regime shift, i.e. the drastic change in the ecosystem organization that has been reported by Alheit et al . (2005 ) and others. These studies exemplify the ability to combine data analytics and domain knowledge through ML to provide explanatory models that provide new insight into ecosystem functioning. Sander et al . (2017 ) used DBNs to infer ecological relationships, but note that presence–absence data may not provide enough signal for these models. Pichler et al . (2020 ) evaluated the ability of multiple ML methods to infer species interactions in the terrestrial domain, but similar approaches could be applied to marine data.

Summarizing ecosystems through regionalization

In recognition that the ocean is spatially and temporally heterogeneous, its division into various types of regions (bioregions, ecoregions, provinces, essential habitats, etc.) provides a means of simplifying and summarizing this heterogeneity into units amenable to further analysis and management. Pioneering this approach was Longhurst et al . (1995 ), who defined 57 biogeochemical provinces mainly using regional variation of remotely sensed chlorophyll-a. In more recent years, ML techniques have been adopted to provide more objective classifications. For example, bioregions have been defined based on chlorophyll-a dynamics using k-means clustering (Mayot et al ., 2016 ) and hierarchical Iso Cluster classification (Welch et al ., 2016 ). Multiple biophysical variables have been used as input to multivariate unsupervised clustering to define pelagic habitats (Hobday, 2011 ; Reygondeau et al ., 2018 ) or track the spatial variability of ocean water masses (Phillips et al ., 2020 ). The concentration of biological organisms derived from survey data (Santora, 2012 ), ecosystem models (Sonnewald et al ., 2020 ), and species distribution models (Welch and McHenry, 2018 ) has also been integrated into classifiers to define ecoregions. Such ecoregions can be useful for spatial planning purposes since they are quite close to the biological targets of such management procedures (Douglass et al ., 2014 ).

Supporting human decisions on ecosystem management

Finally, we also need to evaluate human–ecosystem interactions and define management strategies that support the health and sustainable use of marine ecosystems. These strategies are often defined in intergovernmental texts (e.g. the EU Marine Strategy Framework Directive) that summarize them in terms of quantifiable objectives; ML can help assess those objectives. For example, the likelihood to reach the goals set by the European Union’s Water Framework Directive in Finland was modelled using Bayesian networks (Fernandes et al ., 2012 ). In another example, the accuracy of the automatic classification of plankton images was assessed by checking whether it could provide zooplankton indicators for the EU’s Marine Strategy Framework Directive (Uusitalo et al ., 2016 ).

Early warning regarding specific health indices or potentially harmful species is another area where the fast throughput of ML approaches can improve our practice. A major effort has been spent in predicting algal blooms affecting recreational activities, fisheries, and shellfish farming (Campbell et al ., 2013 ; Fernandes-Salvador et al ., 2021 ). But, similar approaches are used for predicting fish recruitment (Dreyfus-León and Chen, 2007 ; Fernandes et al ., 2010 ) or forecasting litter accumulations on beaches (Granado et al ., 2019 ; Hernández-González et al ., 2019 ).

An international commitment to protect 10% of the ocean by 2020 showcased the importance of spatial planning as a management tool for marine resources (Grorud-Colvert et al ., 2019 ). ML methods, such as automated plankton image classification, are used to monitor and inform the creation of marine protected areas (Muñoz et al ., 2017 ; Benedetti et al ., 2019 ). Dedman et al . ( 2017 ) developed a tool to simplify the use of marine spatial planning tools based on boosted regression trees. Bayesian networks in combination with geographical information systems are being used to analyse conflicting uses, e.g. how to reallocate aquaculture and different fishing fleets with minimal harm (Coccoli et al ., 2018 ; Gimpel et al ., 2018 ), to plan the locations of new activities such as wind energy (Pınarbaşı et al ., 2019 ), or to consider social and economic aspects in addition to environmental ones (Pınarbaşı et al ., 2017 ; Laurila-Pant et al ., 2019 ).

The efficient management of marine ecosystems would require taking decisions that are informed by the current and future states of these systems. ML can be used to build such decision support tools. For example, fish abundance and recruitment are good indicators of the status of fish stocks, and are used to set fishing regulations. But, small pelagic fish recruitment does not follow traditional stock–recruitment relationships, which is why environmental conditions were used to forecast recruitment using ML-based regression (Chen and Ware, 1999 ; Fernandes et al ., 2015 ) and to influence fisheries advice (Fernandes et al ., 2009 ). The ML-based species distribution models described above have been integrated into operational, dynamic, ocean management tools (Hazen et al ., 2018 ; Abrahms et al ., 2019 ), in which management and policy recommendations update regularly in response to changes in biological, environmental, economic, and societal conditions (Welch et al ., 2019 ).

General trends in machine learning applications: data, methods, and tasks

The diversity in the sections above shows that ML is now used in many fields of marine ecology, albeit at different levels of advancement. Several factors can account for the success of the application of ML in a given scientific domain. Based on the examples above, a major one seems to be the type of data: Applications of ML were more successful when they could rely on techniques developed and tested in other fields, which could be repurposed to marine ecology because data were of the same type. This contributes to explaining the disproportionate number of applications of ML to images and videos from cameras, which constitute ∼45% of the references in the database to which ∼15% of references using satellite imagery can be added ( Figure 3 ). Many of those applications benefited from advances in ML motivated by the ubiquity of images in everyday life. For example, several CNN architectures were developed to classify general-purpose image datasets (often ImageNet; Deng et al ., 2009 ), and when they were successful at this task, they also proved relevant for marine applications; for example, the ResNet architecture alone is used in at least 60 papers in the database. Beyond architectures, the weights that result from training CNNs on such large generic datasets are freely distributed by companies (to promote their technology) and can be slightly modified by a short retraining on a marine dataset to yield domain-specific tools (e.g. detect fishes in recordings from underwater cameras). This is called fine-tuning and requires much fewer resources than training from scratch, while yielding very good results. This general approach, called transfer learning, is ubiquitous in the applications of CNNs reviewed above. On the other hand, single-cell spectra obtained from cytometry, for example, constitute a very peculiar type of data and therefore do not benefit from ready-made models; applications of ML to such data are therefore more difficult and scarcer. While sequences of nucleic acids are not common in everyday life, their analysis could still benefit from architectures and pre-trained weights designed for Natural Language Processing, since both are sequences of tokens (e.g. Quang and Xie, 2016 ). However, practically, omics often rely on well-established bioinformatics pipelines, which are not specific to questions in marine ecology and in which some steps do not involve ML; this contributes to explaining the relative scarcity of references from this large field here.

In terms of methods, the four most used algorithms in marine ecological research were, in increasing order of popularity, support vector machines (SVMs), random forests (RFs), convolutional neural networks (CNNs), and non-convolutional artificial neural networks (ANNs; mostly multi-layer perceptrons). ANNs have been used for a long time, which partly explains why they top the list of algorithms; SVMs, and then RFs, came after 2000; since 2013, the usage of CNNs has increased steeply and now they are the ML method most commonly found in new publications ( Figure 7 ). The timing of the usage of those methods in marine ecology largely reflects their appearance or popularization in general: 1995 for SVMs (Cortes and Vapnik, 1995 ), 2001 for RFs (Breiman, 2001 ), and 2012 for CNNs (Krizhevsky et al ., 2012 ); this highlights an early adoption of ML innovations by the marine ecology community. In addition, after the initial adoption, the proportion of studies using them among all studies in marine ecology has grown steeply ( Figure 7 ), which is further evidence of a particular interest for ML approaches in this community. The growth of CNNs, which have progressed the fastest, is associated with their popularity for several data types. Indeed, CNNs take so-called “tensors” as input: multidimensional arrays of numbers. Any type of data that can be made to look like an array within which the proximity between similar numbers is meaningful is amenable to being processed by CNNs. For example, while sounds can be treated as such, most acoustics records can also be represented as spectrograms (intensity as a function of time and frequency), which are tensors and can be processed with models initially designed for images (Stowell, 2022 ). Finally, depending on the output shape and the loss function used, the same network architecture can be used for regression, classification, object detection, etc. (Goodwin et al ., 2022 ).

Amount of references per time period using one of the four most common ML methods in the database. To avoid being misled by the global increase in the number of scientific publications, in any field, the amount is expressed as the proportion of the total number of references published in marine ecology in each time period (defined as the result of the query “WC = (Ecology) AND TS = (marine OR sea OR ocean)” on the Web of Science, i.e. Web of Science category is Ecology and title, abstract, or keywords contain “marine”, “sea”, or “ocean”). All curves increase through time, which means that ML is becoming more common within the field of marine ecology.

Among the papers tagged in the database, ML algorithms are most often used to perform classification (∼60% of references) or regression (∼20%), and, finally, object extraction (detection or segmentation, ∼15%). Yet, the classification of signals, at least, first requires their extraction from the original data (e.g. the detection of an event in a continuous acoustic recording, the segmentation of an organism from an image), so the discrepancy in usage is puzzling. Actually, most automated signal extraction is performed using rules deterministically applied to the raw data. Those rules can be as simple as thresholding (e.g. considering all adjacent dark pixels in an image as objects of interest) but are often much more complex and require both domain expertise to design and signal processing know-how to implement. This hindered the development of automated solutions and explains why objects of interest were (and are still) often extracted manually from underwater videos or acoustics recordings in operational deployments (e.g. Solsona-Berga et al ., 2020 ). DL should enable ecologists to forgo some of the expertise in signal processing and allow extracting signals of interest only from labels placed on a subset of the data. The relative scarcity of their application has likely several explanations. First, deep models for object detection/segmentation are newer (Girshick et al ., 2014 ) than for classification (Lecun et al ., 1998 ) and their applications lag accordingly. Second, they are a bit more complex to set up than classifiers: Drawing bounding boxes or segmentation masks is more time-consuming than sorting files into folders, training classifiers can often start from just this set of sorted raw files, while object detectors/segmenters require text files in a specific format containing the labels linked to the raw data files, etc. However, as labelling tools (e.g. Labelbox), architectures, and reference datasets (e.g. Katija et al ., 2022 ) continue to improve, such applications are likely to explode in the future.

Finally, supervised ML approaches are much more common than unsupervised ones. This is partly linked with the dominance of classification tasks in the references reviewed. Supervised classification is the archetype of task where ML techniques outperform all others: mimic a simple human action, learn it only from examples generated by humans, and be evaluated almost solely on the quality of the prediction.

Limitations for the application of machine learning

Machine learning is particularly effective when the primary concern of ecologists aligns with the performance metric optimized by the technique (e.g. how many images are classified as the correct species?). Conversely, when focus is on both performance and explainability (e.g. how does yearly recruitment intensity depend on environmental variables?), conventional statistics are often chosen over ML. Indeed, ML approaches are commonly qualified as “black boxes”, while people trust models more when they understand the “why” and “how” of their results (Shin, 2021 ). So, when it comes to decision making at least, inherently explainable models are preferred (Rudin, 2019 ). However, those black boxes can be studied, by investigating the importance of each input variable or data point independently through randomization, for example (Lucas, 2020 ). These developments are not unique to ecology and “explainable AI” is an active research domain (Barredo Arrieta et al ., 2020 ).

Another limitation, inherent to ecological data, is the long-tailed distribution of almost everything in the natural world (Preston, 1948 ). Ecosystems are dominated by some species and some processes, yet many others are present at low abundance/frequency and can be key in the response of the system to changes. Such distributions bias the usual loss functions or evaluation metrics (e.g. least-square error in regression, accuracy in classification) and wide data tables (number or variables larger than the number of observations) favour overfitting to the training dataset, which many ML techniques are already prone to. Dealing with imbalanced datasets is a current research topic in ML: In 2020 and 2021, dozens of papers targeted long-tailed distribution and/or imbalance at the Conference on Computer Vision and Pattern Recognition (CVPR), the major conference in the field (e.g. https://openaccess.thecvf.com/CVPR2021 , searching for “long-tail” or “imbalance”). Some of these solutions have been implemented for marine applications, such as rebalancing the training data using data augmentation (Fincham et al ., 2020 ) or generative adversarial networks (GANs; Li et al ., 2021 ), ensembles of several models (Kerr et al ., 2020 ), transfer learning from models trained on balanced data (Lee et al ., 2016 ), etc., but the problem is not solved in the general case.

A consequence of this imbalance is that some of the, numerous, rare classes can largely change in proportion from one data sample to the next, causing a mismatch between the class distribution in the training dataset (usually an average of several samples) and in the new data on which the model will ultimately be applied. This problem, known as “dataset shift” or “concept drift” (Moreno-Torres et al ., 2012 ), is a very common pitfall in the application of ML to marine ecology problems (e.g. Langenkämper et al ., 2020 ) and for the trustworthiness of ML models in general (D’Amour et al ., 2020 ). Indeed, it leads to poor predictive performance that is not necessarily detected when the model is evaluated. Detecting it requires specific validation methods, such as computing evaluation metrics per sample, to capture the inter-sample variability in distribution (Gonzalez et al ., 2017 ). When such a shift is detected, retraining the model with a new training set incorporating more of the natural variability (Langenkämper et al ., 2020 ) or discarding low confidence predictions (Plonus et al ., 2021 ) can help reduce its effect. For classification-type problems, the transition towards quantification approaches, which estimate abundance per class directly, rather than classifying each object, and use the class distribution, can help alleviate it (Gonzalez et al ., 2019 ; Orenstein et al ., 2020 ). Overall, the transferability of a model learned on a given dataset to a dataset with different characteristics is called “domain adaptation” and is also an active field of research (Kouw and Loog, 2019 ).

Finally, ML models are only as good as the datasets they are trained on. Those training datasets are generated by humans, who can make mistakes. For example, in the hard task of discriminating among six dinoflagellate species with large intra-species morphological variations from images, trained scientists achieved 67–83% self-consistency and only 43% consensus (Culverhouse et al ., 2003 ). For the estimation of benthic cover from quadrat pictures in coral reefs, self-consistency of experts ranged from 50 to 90% depending on the type of cover (Beijbom et al ., 2015 ). One upside is that, in both cases, ML models trained on a reference dataset reached performance similar to or higher than human labellers when their inconsistency is taken into account. Another use of ML can be to resolve ambiguous labels and present such potential mistakes to new experts (Schmarje et al ., 2022 ). Finally, a way of alleviating the effect of inconsistent labels is to use additional, independent “ground truth” validation information. For example, in Lekunberri et al . ( 2022 ), estimations of species abundance and size inferred with ML were compared with independent samplings and counting at port when fish were landed. While it is impossible to know which is best between the model-generated or on-the-ground estimates, their discrepancies allow narrowing down on potential biases or difficulties for the experts to accurately label the data.

General outlook

As remarked above, so far, applications of ML in marine ecology have closely followed the development of techniques in computer sciences ( Figure 7 ). However, innovation in ML is accelerating and it may be difficult for marine ecologists to keep track of it. Current developments include transformer-based architectures and diffusion models. Transformers can be seen as an alternative to LSTM recurrent neural networks ( Table 1 ) for sequential inputs, such as language; a well-known everyday application is ChatGPT (​​Chat Generative Pre-Trained Transformer). Their extensions to images are called vision transformers (ViTs), which can be considered as alternatives to CNNs; they have been topping the ImageNet classification challenge since their release (Dosovitskiy et al ., 2021 ). The combination of text and vision models can be used to learn the relationships between images and their captions. New sets of unlabelled images and potential labels can then be placed in the space created by these relationships to label images without any retraining (i.e. zero-shot learning); an operational example is CLIP (Contrastive Language–Image Pre-training). Diffusion models are improved alternatives to generative adversarial networks (GANs) and variational autoencoders to create synthetic images. They can be used to increase the resolution of input images or create completely new images from text input; a popular example is Stable Diffusion.

Now, how could these innovations percolate to marine ecology? Some applications are straightforward. For example, CNNs can simply be swapped for ViTs in image classification tasks to yield better results (Kyathanahally et al ., 2022 ); similarly, GANs could be swapped for diffusion models. Other applications would require more testing: There is no guarantee that the text-to-image relationships learned by CLIP on images from the internet are relevant enough for specific tasks, such as fish species classification from underwater images, for example. Yet ML models have often been discovered to generalize outside their initial domain: Features extracted by a CNN trained on generic images (from ImageNet) were found to be effective for plankton image classification tasks (Orenstein and Beijbom, 2017 ). Still, the question whether the potential improvements brought by these new developments are relevant to solving marine ecology problems remains. Improved performance comes at the cost of larger models (25 M parameters for ResNet50 and 632 M parameters for the ViT-H vision transformer; https://paperswithcode.com/sota/image-classification-on-imagenet ), which require more data to train (Dosovitskiy et al ., 2021 ). Such massive datasets and the computing power to train on them are often only available in large private companies (ChatGPT and CLIP are from OpenAI, the first ViT is from Google, etc.). While the performance benefit is measurable on well-defined challenges (8% increase in accuracy on ImageNet between the two models above), the actual gains on the smaller, noisy, imbalanced datasets of marine ecology, for which global accuracy may not even be a relevant metric, remain to be demonstrated; effort may turn out to be better spent elsewhere.

Actually, the relative performance of existing solutions is already difficult to assess in most subdomains described above because of the lack of standard benchmarks (see also Irisson et al ., 2022 for plankton imaging; and Politikos et al. , 2023 for macrolitter). Such benchmarks depend on the availability of published (and labelled) datasets. The field is progressing on that end, with the release of datasets on e.g. plankton (Sosik et al ., 2015 ) and fish (Fisher et al ., 2016 ) images, remotely sensed images (Kikaki et al ., 2022 ), or ship noise (Santos-Domínguez et al ., 2016 ). However, other datasets, such as images from electronic monitoring on ships, are gathered by private companies that aim to use them to develop and sell electronic monitoring solutions, which are either made mandatory by authorities or desired by fishing companies to reduce costs compared to human observers. For researchers, the effort of gathering and labelling a dataset consistently is often huge and makes some people reluctant to distributing the result openly, although the availability of referenceable repositories (e.g. Zenodo) and citation tracking via Digital Object Identifiers helps. After releasing datasets, the next steps would be to define evaluation metrics following guidelines for proper benchmarking (Weber et al ., 2019 ) and to provide tools to easily track the results of those, now comparable, studies. So, overall, releasing high-quality public datasets, defining benchmarking studies, and centralizing their results are necessary to assess (i) the current state of ML tools in marine ecology and (ii) the tradeoff between gains from new architectures and their cost in complexity.

Transferring innovations from computer sciences to marine ecology also depends largely on efficient collaboration across disciplines. However, establishing interdisciplinary research teams is difficult and takes a long time (Haapasaari et al ., 2012 ). Once again, public datasets are an efficient first step for marine ecologists to garner interest from computer scientists. For example, after the WHOI-Plankton dataset was released (Sosik et al ., 2015 ), it was used in many papers on this topic in computer science conferences. In the assembled database, about 15% of references are from such computer science conferences or engineering journals, but very few are from high-level ones. This can indicate that marine ecology questions have not gained enough interest from the ML research community to generate significant new developments that would be published at high-profile conferences, unlike other applications such as face recognition, customer tracking, or self-driving cars. It could also simply reflect differences in overall funding for research in those fields, linked to the potential commercial applications of some research. On the other hand, publishing highly technical ML papers in ecology journals can also be challenging, because of the scarcity of editors and reviewers who can assess both the importance of the ecological questions and the relevance of the methods used to address them. Still, problems such as estimating the stock sizes of fish species that feed human populations, the distribution of the litter we create, the composition of plankton that forms the basis of oceanic food webs, or the global export of the excess carbon we produce seem no less important than designing targeted ads; they should generate proportionate interest and funding (Blair et al ., 2019 ). Therefore, ways forward for ML in marine ecology include (i) long-term digitalization strategies by funding agencies to scale efforts to the stakes we face and (ii) raising awareness of those stakes among the public in general and computer scientists in particular.

Another way to advance the interplay between ML and marine ecology is to train a new generation of scientists at the intersection of these fields. Then, long-term changes in the strategies for funding allocation and career evaluation would be needed to foster such hybrid profiles. Indeed, garnering the simultaneous interest, on a common problem, of researchers currently specialized in either marine ecology or computer sciences is difficult. Challenges raised by marine ecologists can be perceived as not novel or generic enough to constitute research questions for computer scientists. New developments in computer science are often not immediately actionable by marine ecologists, as seen above for large transformers or image-text encoders. So, computer scientists may feel like service providers for ecologists and ecologists as simple data providers for computer scientists, which is satisfactory for neither. Recent reviews and perspectives, in ecology as a whole, actually show that the interaction can be beneficial for both parties. Several also point towards the need to train ecologists in computer sciences, not the opposite (Olden et al ., 2008 for an older one; Christin et al ., 2019 for a recent one), notably because computer science students rarely choose careers in ecology and environment, in part due to differences in financial compensation or job security. Overall, we argue that interdisciplinary training and career paths are potential solutions to many of the current shortcomings of ML applications in marine ecology.

All authors acknowledge the support of ICES through the Working group on Machine Learning in Marine Science (WGMLEARN).

No conflict of interest was reported.

AP, AG, BK, CJ, CP, DSG, DP, DDB, HM, JBR, JOI, JF, JTW, JAF, KM, KOM, KHK, LU, LVdB, ML, ODBP, PR, PS, RS, RML, SC, TC, and WM built and labelled the publications database. Contributions are then broken down by section, listing the main providers of content: section “What is machine learning and why does marine ecology need it?”—HM, JOI, KM, PR, and RK; section “A quick primer on machine learning”—JOI, KM, PR, and VS; section “The setup of the database and its tags”—JOI and PR; section “Benthos” AG, JF, and KH; section “Macrolitter”—DP and SV; section “Nekton”—HM, JBR, MS, and ATMvH; section “Plankton”—HM, JOI, KOM, and RK; section “Identifying microorganisms from single-cell spectra”—PR; section “Active acoustics for target classification”—AP, HM, and NOH; section “Active acoustics for seabed and sediment mapping“—KHK; section “Passive acoustics monitoring”—CP, DDB, PD, and VS; section “Profiling biological communities with environmental genomics”—HM, KM, LVdB, SC, and TC; section “Quantifying and mapping fishing pressure from geolocation data”—JAF; section “Deriving biogeochemical variables from satellite images and floats profiles”—CJ and RS; section “Predicting species abundance and distribution”—SB and HW; section “Capturing dynamic ecological relationships”—LU and JAF; section “Summarizing ecosystems through regionalization”—SB and HW; section “Supporting human decisions on ecosystem management”—JAF; and section “Discussion and perspectives”—JOI. In addition, DSG, DP, HM, HW, JOI, JTW, JAF, ODBP, PR, RK, RML, SC, and SB reviewed the whole text. PR, KM, and JOI led the overall project.

TC acknowledges support from the Swiss National Science Foundation (#31003A_179125), the European Research Council (#818449 AGENSI), and the Horizon Europe programme (#101094924 ANERIS). JAF-S has received funding from the project H2020 FutureMARES (#869300) and SusTunTech (#869342). NOH acknowledges support from the CRIMAC centre funded by the Research Council of Norway #309512. KH is supported by the Mission Atlantic project funded by the European Union’s Horizon 2020 Research and Innovation Programme (#862428). MS acknowledges funding from the European Union’s H2020 programme #7553521 (SMARTFISH); European’s Maritime and Fisheries Fund and the Danish Fisheries Agency, #33112-I-19-076 (AutoCatch); and Fully Documented Fisheries, funded by the European Maritime and Fisheries Fund (EMFF). LVdB acknowledges support from the Sand Fund of the Federal Public Service Economy. SC, RML, and SV acknowledge support from the H2020 project AtlantECO (#862923). RML acknowledges support from CNPq, Brazil (grant number 315033/2021-5). RK acknowledges support via a “Make Our Planet Great Again” grant of the French National Research Agency within the “Programme d’Investissements d’Avenir” (#ANR-19-MPGA-0012), by the Heisenberg programme of the German Science Foundation (#469175784), and from NOAA (#NA21OAR4310254). JBR acknowledges funding from the IFREMER Scientific Direction project DEEP. KM’s participation was funded by the Norwegian Ministry of Trade, Industry and Fisheries. JOI acknowledges funding from the Belmont Forum project WWWPIC (#ANR-018-BELM-0003–01).

The data underlying this article are available in the article and in its online supplementary material .

Abadi   M. , Agarwal   A. , Barham   P. , Brevdo   E. , Chen   Z. , Citro   C. , Corrado   G. S  et al.    2016 . TensorFlow: large-scale machine learning on heterogeneous distributed systems . http://arxiv.org/abs/1603.04467 .

Abrahms   B. , Welch   H. , Brodie   S. , Jacox   M. G. , Becker   E. A. , Bograd   S. J. , Irvine   L. M  et al.    2019 . Dynamic ensemble models to predict distributions and anthropogenic risk exposure for highly mobile species . Diversity and Distributions , 25 : 1182 – 1193 .

Google Scholar

Alheit   J. , Möllmann   C. , Dutz   J. , Kornilovs   G. , Loewe   P. , Mohrholz   V. , Wasmund   N.   2005 . Synchronous ecological regime shifts in the central Baltic and the North Sea in the late 1980s . ICES Journal of Marine Science , 62 : 1205 – 1215 .

Allen   A. S. , Yurk   H. , Vagle   S. , Pilkington   J. , Canessa   R.   2018 . The underwater acoustic environment at SGaan Kinghlas–Bowie seamount marine protected area: characterizing vessel traffic and associated noise using satellite AIS and acoustic datasets . Marine Pollution Bulletin , 128 : 82 – 88 .

Allken   V. , Handegard   N. O. , Rosen   S. , Schreyeck   T. , Mahiout   T. , Malde   K.   2019 . Fish species identification using a convolutional neural network trained on synthetic data . ICES Journal of Marine Science , 76 : 342 – 349 .

Althaus   F. , Hill   N. , Ferrari   R. , Edwards   L. , Przeslawski   R. , Schönberg   C. H. L. , Stuart-Smith   R  et al.    2015 . A standardised vocabulary for identifying benthic biota and substrata from underwater imagery: the CATAMI classification scheme . PLoS One , 10 : e0141039 .

Anderson   J. T. , Holliday   D. V. , Kloser   R. , Reid   D. G. , Simard   Y.   2008 . Acoustic seabed classification: current practice and future directions . ICES Journal of Marine Science , 65 : 1004 – 1011 .

Baker   R. E. , Peña   J.-M. , Jayamohan   J. , Jérusalem   A.   2018 . Mechanistic models versus machine learning, a fight worth fighting for the biological community? . Biology Letters , 14 : 20170660 .

Barredo Arrieta   A. , Díaz-Rodríguez   N. , Del Ser   J. , Bennetot   A. , Tabik   S. , Barbado   A. , Garcia   S  et al.    2020 . Explainable artificial intelligence (XAI): concepts, taxonomies, opportunities and challenges toward responsible AI . Information Fusion , 58 : 82 – 115 .

Beale   C. M. , Lennon   J. J.   2012 . Incorporating uncertainty in predictive species distribution modelling . Philosophical Transactions of the Royal Society B: Biological Sciences , 367 : 247 – 258 .

Behivoke   F. , Etienne   M.-P. , Guitton   J. , Randriatsara   R. M. , Ranaivoson   E. , Léopold   M.   2021 . Estimating fishing effort in small-scale fisheries using GPS tracking data and random forests . Ecological Indicators , 123 : 107321 .

Beijbom   O. , Edmunds   P. J. , Kline   D. I. , Mitchell   B. G. , Kriegman   D.   2012 . Automated annotation of coral reef survey images . IEEE Conference on Computer Vision and Pattern Recognition , Providence , RI , pp. 1170 – 1177 .

Beijbom   O. , Edmunds   P. J. , Roelfsema   C. , Smith   J. , Kline   D. I. , Neal   B. P. , Dunlap   M. J  et al.    2015 . Towards automated annotation of benthic survey images: variability of human experts and operational modes of automation . PLoS One , 10 : e0130312 .

Benedetti   F. , Jalabert   L. , Sourisseau   M. , Beker   B. , Cailliau   C. , Desnos   C. , Elineau   A  et al.    2019 . The seasonal and inter-annual fluctuations of plankton abundance and community structure in a North Atlantic Marine Protected Area . Frontiers in Marine Science , 6 : 214 .

Benfield   M. C. , Grosjean   P. , Culverhouse   P. F. , Irigoien   X. , Sieracki   M. E. , Lopez-Urrutia   A. , Dam   H. G  et al.    2007 . RAPID: research on automated plankton identification . Oceanography , 20 : 172 – 187 .

Benkendorf   D.   2020 . Effects of sample size and network depth on a deep learning approach to species distribution modeling . Ecological Informatics , 60 : 101137 .

Bergler   C. , Schmitt   M. , Cheng   R. X. , Schröter   H. , Maier   A. , Barth   V. , Weber   M  et al.    2019 . Deep representation learning for orca call type classification . In   Text, Speech, and Dialogue , pp. 274 – 286 .. Ed. by   Ekštein   K. . Springer International Publishing , Cham .

Google Preview

Bermant   P. C . 2021 . BioCPPNet: automatic bioacoustic source separation with deep neural networks . Scientific Reports , 11 : 23502 .

Bermant   P. C. , Bronstein   M. M. , Wood   R. J. , Gero   S. , Gruber   D. F.   2019 . Deep machine learning techniques for the detection and classification of sperm whale bioacoustics . Scientific Reports , 9 : 12588 .

Bertrand   S. , Díaz   E. , Lengaigne   M.   2008 . Patterns in the spatial distribution of peruvian anchovy ( Engraulis ringens ) revealed by spatially explicit fishing data . Progress in Oceanography , 79 : 379 – 389 .

Beslin   W. A. M. , Whitehead   H. , Gero   S . 2018 . Automatic acoustic estimation of sperm whale size distributions achieved through machine recognition of on-axis clicks . The Journal of the Acoustical Society of America , 144 : 3485 – 3495 .

Beyan   C. , Browman   H. I.   2020 . Setting the stage for the machine intelligence era in marine science . ICES Journal of Marine Science , 77 : 1267 – 1273 .

Bittig   H. C. , Steinhoff   T. , Claustre   H. , Fiedler   B. , Williams   N. L. , Sauzède   R. , Körtzinger   A  et al.    2018 . An alternative to static climatologies: robust estimation of open ocean CO 2 variables and nutrient concentrations from T, S, and O 2 data using Bayesian neural networks . Frontiers in Marine Science , 5 : 328 .

Bittle   M. , Duncan   A . 2013 . A review of current marine mammal detection and classification algorithms for use in automated passive acoustic monitoring . In   Proceedings of Acoustics: Science, Technology and Amenity , pp. 1 – 8 .. Ed. by McMinn   T. , Australian Acoustical Society , Victor Harbour, South Australia .

Blair   G. S. , Henrys   P. , Leeson   A. , Watkins   J. , Eastoe   E. , Jarvis   S. , Young   P. J.   2019 . Data science of the natural environment: a research roadmap . Frontiers in Environmental Science , 7 : 121.

Bochinski   E. , Bacha   G. , Eiselein   V. , Walles   T. J. W. , Nejstgaard   J. C. , Sikora   T.   2019 . Deep active learning for in situ plankton classification . In   Pattern Recognition and Information Forensics , pp. 5 – 15 .. Ed. by   Zhang   Z. , Suter   D. , Tian   Y. , Branzan Albu   A. , Sidère   N. , Escalante   H. Jair . Springer International Publishing , Cham .

Boddy   L. , Wilkins   M. F. , Morris   C. W.   2001 . Pattern recognition in flow cytometry . Cytometry , 44 : 195 – 209 .

Bogucki   R. , Cygan   M. , Khan   C. B. , Klimek   M. , Milczek   J. K. , Mucha   M.   2019 . Applying deep learning to right whale photo identification . Conservation Biology , 33 : 676 – 684 .

Bohan   D. A. , Vacher   C. , Tamaddoni-Nezhad   A. , Raybould   A. , Dumbrell   A. J. , Woodward   G.   2017 . Next-generation global biomonitoring: large-scale, automated reconstruction of ecological networks . Trends in Ecology and Evolution , 32 : 477 – 487 .

Bolyen   E. , Rideout   J. R. , Dillon   M. R. , Bokulich   N. A. , Abnet   C. C. , Al-Ghalith   G. A. , Alexander   H  et al.    2019 . Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2 . Nature biotechnology , 37 : 852 – 857 .

Boulais   O. , Woodward   B. , Schlining   B. , Lundsten   L. , Barnard   K. , Bell   K. C. , Katija   K.   2020 . FathomNet: an underwater image training database for ocean exploration and discovery . http://arxiv.org/abs/2007.00114 .

Bowman   J. S. , Amaral-Zettler   L. A. , J Rich   J. , M Luria   C. , Ducklow   H. W . 2017 . Bacterial community segmentation facilitates the prediction of ecosystem function along the coast of the western Antarctic Peninsula . The ISME Journal , 11 : 1460 – 1471 .

Brajard   J. , Santer   R. , Crépon   M. , Thiria   S.   2012 . Atmospheric correction of MERIS data for case-2 waters using a neuro-variational inversion . Remote Sensing of Environment , 126 : 51 – 61 .

Brautaset   O. , Waldeland   A. U. , Johnsen   E. , Malde   K. , Eikvil   L. , Salberg   A.-B. , Handegard   N. O.   2020 . Acoustic classification in multifrequency echosounder data using deep convolutional neural networks . ICES Journal of Marine Science , 77 : 1391 – 1400 .

Breiman   L.   2001 . Random forests . Machine learning , 45 : 5 – 32 .

Brodie   S. J. , Thorson   J. T. , Carroll   G. , Hazen   E. L. , Bograd   S. , Haltuch   M. A. , Holsman   K. K  et al.    2020 . Trade-offs in covariate selection for species distribution models: a methodological comparison . Ecography , 43 : 11 – 24 .

Brodie   S. , Jacox   M. G. , Bograd   S. J. , Welch   H. , Dewar   H. , Scales   K. L. , Maxwell   S. M  et al.    2018 . Integrating dynamic subsurface habitat metrics into species distribution models . Frontiers in Marine Science , 5 : 219.

Brown   C. J. , Smith   S. J. , Lawton   P. , Anderson   J. T.   2011 . Benthic habitat mapping: a review of progress towards improved understanding of the spatial ecology of the seafloor using acoustic techniques . Estuarine, Coastal and Shelf Science , 92 : 502 – 520 .

Brown   J. C. , Smaragdis   P. , Nousek-McGregor   A.   2010 . Automatic identification of individual killer whales . The Journal of the Acoustical Society of America , 128 : EL93 – EL98 .

Bruckmann   C. , Müller   S. , Höner zu Siederdissen   C . 2022 . Automatic, fast, hierarchical, and non-overlapping gating of flow cytometric data with flowEMMi v2 . Computational and Structural Biotechnology Journal , 20 : 6473 – 6489 .

Cabreira   A. G. , Tripode   M. , Madirolas   A . 2009 . Artificial neural networks for fish-species identification . ICES Journal of Marine Science , 66 : 1119 – 1129 .

Callahan   B. J. , McMurdie   P. J. , Rosen   M. J. , Han   A. W. , Johnson   A. J. A. , Holmes   S. P . 2016 . DADA2: high-resolution sample inference from Illumina amplicon data . Nature Methods , 13 : 581 – 583 .

Campbell   L. , Henrichs   D. W. , Olson   R. J. , Sosik   H. M . 2013 . Continuous automated imaging-in-flow cytometry for detection and early warning of Karenia brevis blooms in the Gulf of Mexico . Environmental Science and Pollution Research , 20 : 6896 – 6902 .

Canals   M. , Pham   C. K. , Bergmann   M. , Gutow   L. , Hanke   G. , van Sebille   E. , Angiolillo   M  et al.    2020 . The quest for seafloor macrolitter: a critical review of background knowledge, current methods and future prospects . Environmental Research Letters , 16 ( 2 ): 023001 .

Carlson   C. J.   2020 . embarcadero: species distribution modelling with Bayesian additive regression trees in R . Methods in Ecology and Evolution , 11 : 850 – 858 .

Chaffron   S. , Delage   E. , Budinich   M. , Vintache   D. , Henry   N. , Nef   C. , Ardyna   M  et al.    2021 . Environmental vulnerability of the global ocean epipelagic plankton community interactome . Science Advances , 7 : eabg1921 .

Chen   D. G. , Ware   D. M.   1999 . A neural network model for forecasting fish stock recruitment . Canadian Journal of Fisheries and Aquatic Sciences , 56 : 2385 .

Choi   C. , Kampffmeyer   M. , Handegard   N. O. , Salberg   A. , Brautaset   O. , Eikvil   L. , Jenssen   R.   2021 . Semi-supervised target classification in multi-frequency echosounder data . ICES Journal of Marine Science , 78 : 2615 – 2627 .,

Christin   S. , Hervet   É. , Lecomte   N.   2019 . Applications for deep learning in ecology . Methods in Ecology and Evolution , 10 : 1632 – 1644 .

Cimino   M. A. , Santora   J. A. , Schroeder   I. , Sydeman   W. , Jacox   M. G. , Hazen   E. L. , Bograd   S. J.   2020 . Essential krill species habitat resolved by seasonal upwelling and ocean circulation models within the large marine ecosystem of the California Current System . Ecography , 43 : 1536 – 1549 .

Claustre   H. , Johnson   K. S. , Takeshita   Y.   2020 . Observing the global ocean with Biogeochemical-Argo . Annual review of marine science , 12 : 23 – 48 .

Coccoli   C. , Galparsoro   I. , Murillas   A. , Pınarbaşı   K. , Fernandes   J. A.   2018 . Conflict analysis and reallocation opportunities in the framework of marine spatial planning: a novel, spatially explicit Bayesian belief network approach for artisanal fishing and aquaculture . Marine Policy , 94 : 119 – 131 .

Cordier   T. , Forster   D. , Dufresne   Y. , Martins   C. I. M. , Stoeck   T. , Pawlowski   J.   2018 . Supervised machine learning outperforms taxonomy-based environmental DNA metabarcoding applied to biomonitoring . Molecular Ecology Resources , 18 : 1381 – 1391 .

Cortes   C. , Vapnik   V.   1995 . Support vector machine . Machine learning , 20 : 273 – 297 .

Crespo   G. O. , Dunn   D. C. , Reygondeau   G. , Boerder   K. , Worm   B. , Cheung   W. , Tittensor   D. P  et al.    2018 . The environmental niche of the global high seas pelagic longline fleet . Science Advances , 4 : eaat3681 .

Cui   X. , Yang   F. , Wang   X. , Ai   B. , Luo   Y. , Ma   D.   2021 . Deep learning model for seabed sediment classification based on fuzzy ranking feature optimization . Marine Geology , 432 : 106390 .

Culverhouse   P. F. , Williams   R. , Benfield   M. , Flood   P. R. , Sell   A. F. , Mazzocchi   M. G. , Buttino   I  et al.    2006 . Automatic image analysis of plankton: future perspectives . Marine Ecology Progress Series , 312 : 297 – 309 .

Culverhouse   P. F. , Williams   R. , Reguera   B. , Herry   V. , González-Gil   S.   2003 . Do experts make mistakes? A comparison of human and machine indentification of dinoflagellates . Marine Ecology Progress Series , 247 : 17 – 25 .

D'Amour   A. , Heller   K. , Moldovan   D. , Adlam   B. , Alipanahi   B. , Beutel   A. , Chen   C. , et al.   2020 . Underspecification presents challenges for credibility in modern machine learning . http://arxiv.org/abs/2011.03395 .

Dartnell   P. , Gardner   J. V.   2004 . Predicting seafloor facies from multibeam bathymetry and backscatter data . Photogrammetric Engineering and Remote Sensing , 70 : 1081 – 1091 .

Dedman   S. , Officer   R. , Clarke   M. , Reid   D. G. , Brophy   D.   2017 . Gbm.auto: a software tool to simplify spatial modelling and marine protected area planning . PLoS One , 12 : e0188955 .

Deneu   B. , Servajean   M. , Bonnet   P. , Botella   C. , Munoz   F. , Joly   A.   2021 . Convolutional neural networks improve species distribution modelling by capturing the spatial structure of the environment . PLoS Computational Biology , 17 : e1008856 .

Deng   J. , Dong   W. , Socher   R. , Li   L.-J. , Li   K. , Fei-Fei   L.   2009 . ImageNet: a large-scale hierarchical image database . In   IEEE Conference on Computer Vision and Pattern Recognition , pp. 248 – 255 .. IEEE .

Derville   S. , Torres   L. G. , Iovan   C. , Garrigue   C.   2018 . Finding the right fit: comparative cetacean distribution models using multiple data sources and statistical approaches . Diversity and Distributions , 24 : 1657 – 1673 .

Dosovitskiy   A. , Beyer   L. , Kolesnikov   A. , Weissenborn   D. , Zhai   X. , Unterthiner   T. , Dehghani   M  et al.    2021 , An image is worth 16 × 16 words: transformers for image recognition at scale . http://arxiv.org/abs/2010.11929 .

Douglass   L. L. , Turner   J. , Grantham   H. S. , Kaiser   S. , Constable   A. , Nicoll   R. , Raymond   B  et al.    2014 . A hierarchical classification of benthic biodiversity and assessment of protected areas in the Southern Ocean . PLoS One , 9 : e100551 .

Dreyfus-León   M. , Chen   D. G.   2007 . Recruitment prediction with genetic algorithms with application to the Pacific Herring fishery . Ecological Modelling , 203 : 141 – 146 .

Dreyfus-Leon   M. , Kleiber   P.   2001 . A spatial individual behaviour-based model approach of the yellowfin tuna fishery in the eastern Pacific Ocean . Ecological Modelling , 146 : 47 – 56 .

Ducharme-Barth   N. D. , Ahrens   R. N. M.   2017 . Classification and analysis of VMS data in vertical line fisheries: incorporating uncertainty into spatial distributions . Canadian Journal of Fisheries and Aquatic Sciences , 74 : 1749 – 1764 .

Durden   J. M. , Hosking   B. , Bett   B. J. , Cline   D. , Ruhl   H. A.   2021 . Automated classification of fauna in seabed photographs: the impact of training and validation dataset size, with considerations for the class imbalance . Progress in Oceanography , 196 : 102612 .

Elith   J. , Leathwick   J. R.   2009 . Species distribution models: ecological explanation and prediction across space and time . Annual Review of Ecology, Evolution, and Systematics , 40 : 677 – 697 .

Elith   J. , Leathwick   J. R. , Hastie   T.   2008 . A working guide to boosted regression trees . Journal of Animal Ecology , 77 : 802 – 813 .

Elith   J. , Phillips   S. J. , Hastie   T. , Dudík   M. , Chee   Y. E. , Yates   C. J.   2011 . A statistical explanation of MaxEnt for ecologists: statistical explanation of MaxEnt . Diversity and Distributions , 17 : 43 – 57 .

Ellen   J. S. , Graff   C. A. , Ohman   M. D.   2019 . Improving plankton image classification using context metadata . Limnology and Oceanography: Methods , 17 : 439 – 461 .

Fallon   N. G. , Fielding   S. , Fernandes   P. G.   2016 . Classification of Southern Ocean krill and icefish echoes using random forests . ICES Journal of Marine Science , 73 : 1998 – 2008 .

Faust   K. , Raes   J.   2012 . Microbial interactions: from networks to models . Nature Reviews Microbiology , 10 : 538 – 550 .

Fernandes   J. A. , Irigoien   X. , Goikoetxea   N. , Lozano   J. A. , Inza   I. , Pérez   A. , Bode   A.   2010 . Fish recruitment prediction, using robust supervised classification methods . Ecological Modelling , 221 : 338 – 352 .

Fernandes   J. A. , Irigoien   X. , Lozano   J. A. , Inza   I. , Goikoetxea   N. , Pérez   A.   2015 . Evaluating machine-learning techniques for recruitment forecasting of seven North East Atlantic fish species . Ecological Informatics , 25 : 35 – 42 .

Fernandes   J. A. , Kauppila   P. , Uusitalo   L. , Fleming-Lehtinen   V. , Kuikka   S. , Pitkänen   H.   2012 . Evaluation of reaching the targets of the water framework directive in the Gulf of Finland . Environmental Science and Technology , 46 : 8220 – 8228 .

Fernandes   J. A. , Lozano   J. A. , Inza   I. , Irigoien   X. , Pérez   A. , Rodríguez   J. D.   2013 . Supervised pre-processing approaches in multiple class variables classification for fish recruitment forecasting . Environmental Modelling and Software , 40 : 245 – 254 .

Fernandes   J. , Irigoien   X. , Uriarte   A. , Ibaibarriaga   L. , Lozano   J. , Inza   I.   2009 . Anchovy recruitment mixed long series prediction using supervised classification . Working document to the ICES Benchmark Workshop on Short-lived Species (WKSHORT) .

Fernandes-Salvador   J. A. , Davidson   K. , Sourisseau   M. , Revilla   M. , Schmidt   W. , Clarke   D. , Miller   P. I ., et al.    2021 . Current status of forecasting toxic harmful algae for the north-east Atlantic shellfish aquaculture industry . Frontiers in Marine Science , 8 : 666583 .

Fincham   J. I. , Wilson   C. , Barry   J. , Bolam   S. , French   G.   2020 . Developing the use of convolutional neural networking in benthic habitat classification and species distribution modelling . ICES Journal of Marine Science , 77 : 3074 – 3082 .

Fisher   R. B. , Chen-Burger   Y.-H. , Giordano   D. , Hardman   L. , Lin   F.-P. (Eds). 2016 . Fish4Knowledge: Collecting and Analyzing Massive Coral Reef Fish Video Data. Intelligent Systems Reference Library . Springer International Publishing , Cham .

Flück   B. , Mathon   L. , Manel   S. , Valentini   A. , Dejean   T. , Albouy   C. , Mouillot   D  et al.    2022 . Applying convolutional neural networks to speed up environmental DNA annotation in a highly diverse ecosystem . Scientific Reports , 12 : 10247 .

Fournier   M. , Casey Hilliard   R. , Rezaee   S. , Pelot   R.   2018 . Past, present, and future of the satellite-based automatic identification system: areas of applications (2004–2016) . WMU Journal of Maritime Affairs , 17 : 311 – 345 .

Friedman   J. , Hastie   T. , Tibshirani   R.   2001 . The Elements of Statistical Learning . Springer , New York, NY .

Frühe   L. , Cordier   T. , Dully   V. , Breiner   H.-W. , Lentendu   G. , Pawlowski   J. , Martins   C  et al.    2021 . Supervised machine learning is superior to indicator value inference in monitoring the environmental impacts of salmon aquaculture using eDNA metabarcodes . Molecular Ecology , 30 : 2988 – 3006 .

Gallager   S. M.   2019 . System for rapid assessment of water quality and harmful algal bloom toxins . US patent: UW 2019/0293565 A1 .

Garcia   R. , Prados   R. , Quintana   J. , Tempelaar   A. , Gracias   N. , Rosen   S. , Vågstøl   H  et al.    2020 . Automatic segmentation of fish using deep learning with application to fish size measurement . ICES Journal of Marine Science , 77 : 1354 – 1366 .

Garcia-Garin   O. , Monleón-Getino   T. , López-Brosa   P. , Borrell   A. , Aguilar   A. , Borja-Robalino   R. , Cardona   L  et al.    2021 . Automatic detection and quantification of floating marine macro-litter in aerial images: introducing a novel deep learning approach connected to a web application in R . Environmental Pollution , 273 : 116490 .

Gibb   R. , Browning   E. , Glover-Kapfer   P. , Jones   K. E.   2019 . Emerging opportunities and challenges for passive acoustics in ecological assessment and monitoring . Methods in Ecology and Evolution , 10 : 169 – 185 .

Gimpel   A. , Stelzenmüller   V. , Töpsch   S. , Galparsoro   I. , Gubbins   M. , Miller   D. , Murillas   A  et al.    2018 . A GIS-based tool for an integrated assessment of spatial planning trade-offs with aquaculture . Science of the Total Environment , 627 : 1644 – 1655 .

Girshick   R. , Donahue   J. , Darrell   T. , Malik   J.   2014 . Rich feature hierarchies for accurate object detection and semantic segmentation . In IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , IEEE , Columbus, OH , pp. 580 – 587 .

Gómez-Ríos   A. , Tabik   S. , Luengo   J. , Shihavuddin   A. , Krawczyk   B. , Herrera   F.   2019 . Towards highly accurate coral texture images classification using deep convolutional neural networks and data augmentation . Expert Systems with Applications , 118 : 315 – 328 .

Gonzalez   P. , Alvarez   E. , Diez   J. , Lopez-Urrutia   A. , del Coz   J. J.   2017 . Validation methods for plankton image classification systems . Limnology and Oceanography: Methods , 15 : 221 – 237 .

Gonzalez   P. , Castano   A. , Peacock   E. E. , Diez   J. , Jose Del Coz   J. , Sosik   H. M.   2019 . Automatic plankton quantification using deep features . Journal of Plankton Research , 41 : 449 – 463 .

Goodwin   M. , Halvorsen   K. T. , Jiao   L. , Knausgård   K. M. , Martin   A. H. , Moyano   M. , Oomen   R. A  et al.    2022 . Unlocking the potential of deep learning for marine ecology: overview, applications, and outlook . ICES Journal of Marine Science , 79 : 319 – 336 .

Gorsky   G. , Ohman   M. D. , Picheral   M. , Gasparini   S. , Stemmann   L. , Romagnan   J.-B. , Cawood   A  et al.    2010 . Digital zooplankton image analysis using the ZooScan integrated system . Journal of Plankton Research , 32 : 285 – 303 .

Granado   I. , Basurko   O. C. , Rubio   A. , Ferrer   L. , Hernández-González   J. , Epelde   I. , Fernandes   J. A.   2019 . Beach litter forecasting on the south-eastern coast of the Bay of Biscay: a Bayesian networks approach . Continental Shelf Research , 180 : 14 – 23 .

Grorud-Colvert   K. , Constant   V. , Sullivan-Stack   J. , Dziedzic   K. , Hamilton   S. L. , Randell   Z. , Fulton-Bennett   H  et al.    2019 . High-profile international commitments for ocean protection: empty promises or meaningful progress? . Marine Policy , 105 : 52 – 66 .

Grosjean   P. , Picheral   M. , Warembourg   C. , Gorsky   G.   2004 . Enumeration, measurement, and identification of net zooplankton samples using the ZOOSCAN digital imaging system . ICES Journal of Marine Science , 61 : 518 – 525 .

Gugele   S. M. , Widmer   M. , Baer   J. , DeWeber   J. T. , Balk   H. , Brinker   A.   2021 . Differentiation of two swim bladdered fish species using next generation wideband hydroacoustics . Scientific Reports , 11 : 10520 .

Guidi   L. , Chaffron   S. , Bittner   L. , Eveillard   D. , Larhlimi   A. , Roux   S. , Darzi   Y  et al.    2016 . Plankton networks driving carbon export in the oligotrophic ocean . Nature , 532 : 465 – 470 .

Guidi   L. , Fernàndez-Guerra   A. , Canchaya   C. , Curry   E. , Foglini   F. , Irisson   J.-O. , Malde   K  et al.    2020 . Big data in marine science . In Future Science Brief 6 of the European Marine Board . Ed. by Alexander   B. , Heymans   J. J. , Muñiz Piniella   A. , Kellett   P. , Coopman   J. . European Marine Board , Ostend .

Guillou   L. , Bachar   D. , Audic   S. , Bass   D. , Berney   C. , Bittner   L. , Boutte   C  et al.    2013 . The Protist Ribosomal Reference database (PR2): a catalog of unicellular eukaryote small sub-unit rRNA sequences with curated taxonomy . Nucleic Acids Research , 41 : D597 – D604 .

Guisan   A. , Thuiller   W.   2005 . Predicting species distribution: offering more than simple habitat models . Ecology Letters , 8 : 993 – 1009 .

Haapasaari   P. , Kulmala   S. , Kuikka   S.   2012 . Growing into interdisciplinarity: how to converge biology, economics, and social science in fisheries research? . Ecology and Society , 17 : 6 .

Hamilton   L. J.   2001 . Acoustic seabed classification systems . Report no. DSTO-TN-0401, Defence Science and Technolgy Organisation, Aeronautical and Martime Research Lab .

Hazen   E. L. , Scales   K. L. , Maxwell   S. M. , Briscoe   D. K. , Welch   H. , Bograd   S. J. , Bailey   H  et al.    2018 . A dynamic ocean management tool to reduce bycatch and support sustainable fisheries . Science Advances , 4 : eaar3001 .

Helmond   A. T. M. , Mortensen   L. O. , Plet-Hansen   K. S. , Ulrich   C. , Needle   C. L. , Oesterwind   D. , Kindt-Larsen   L  et al.    2020 . Electronic monitoring in fisheries: lessons from global experiences and future opportunities . Fish and Fisheries , 21 : 162 – 189 .

Hernández-González   J. , Inza   I. , Granado   I. , Basurko   O. C. , Fernandes   J. A. , Lozano   J. A.   2019 . Aggregated outputs by linear models: an application on marine litter beaching prediction . Information Sciences , 481 : 381 – 393 .

Hindell   M. A. , Reisinger   R. R. , Ropert-Coudert   Y. , Hückstädt   L. A. , Trathan   P. N. , Bornemann   H. , Charrassin   J.-B  et al.    2020 . Tracking of marine predators to protect Southern Ocean ecosystems . Nature , 580 : 87 – 92 .

Hirama   Y. , Yokoyama   S. , Yamashita   T. , Kawamura   H. , Suzuki   K. , Wada   M.   2017 . Discriminating fish species by an echo sounder in a set-net using a CNN . 21st Asia Pacific Symposium on Intelligent and Evolutionary Systems (IES) , Hanoi , pp. 112 – 115 .

Hobday   A.   2011 . Defining dynamic pelagic habitats in oceanic waters off eastern Australia . Deep Sea Research Part II: Topical Studies in Oceanography , 58 : 734 – 745 .

Howell   K. L. , Davies   J. S. , Allcock   A. L. , Braga-Henriques   A. , Buhl-Mortensen   P. , Carreiro-Silva   M. , Dominguez-Carrió   C  et al.    2019 . A framework for the development of a global standardised marine taxon reference image database (SMarTaR-ID) to support image-based analyses . PLoS One , 14 : e0218904 .

Hu   J. , Li   D. , Duan   Q. , Han   Y. , Chen   G. , Si   X.   2012 . Fish species classification by color, texture and multi-class support vector machine using computer vision . Computers and Electronics in Agriculture , 88 : 133 – 140 .

Hu   Q. , Davis   C.   2005 . Automatic plankton image recognition with co-occurrence matrices and support vector machine . Marine Ecology Progress Series , 295 : 21 – 31 .

Hunt   T. N. , Allen   S. J. , Bejder   L. , Parra   G. J.   2020 . Identifying priority habitat for conservation and management of Australian humpback dolphins within a marine protected area . Scientific Reports , 10 : 14366 .

Hyrkas   J. , Clayton   S. , Ribalet   F. , Halperin   D. , Virginia Armbrust   E. , Howe   B.   2016 . Scalable clustering algorithms for continuous environmental flow cytometry . Bioinformatics , 32 : 417 – 423 .

Ierodiaconou   D. , Monk   J. , Rattray   A. , Laurenson   L. , Versace   V. L.   2011 . Comparison of automated classification techniques for predicting benthic biological communities using hydroacoustics and video observations . Continental Shelf Research , 31 : S28 – S38 .

Inada   K. , Matsuda   R. , Fujiwara   C. , Nomura   M. , Tamon   T. , Nishihara   I. , Takao   T  et al.    2001 . Identification of plastics by infrared absorption using InGaAsP laser diode . Resources, Conservation and Recycling , 33 : 131 – 146 .

Ioannou   I. , Gilerson   A. , Gross   B. , Moshary   F. , Ahmed   S.   2011 . Neural network approach to retrieve the inherent optical properties of the ocean from observations of MODIS . Applied Optics , 50 : 3168 .

Ioannou   I. , Gilerson   A. , Gross   B. , Moshary   F. , Ahmed   S.   2013 . Deriving ocean color products using neural networks . Remote Sensing of Environment , 134 : 78 – 91 .

Irigoien   X. , Fernandes   J. A. , Grosjean   P. , Denis   K. , Albaina   A. , Santos   M.   2009 . Spring zooplankton distribution in the Bay of Biscay from 1998 to 2006 in relation with anchovy recruitment . Journal of Plankton Research , 31 : 1 – 17 .

Irisson   J.-O. , Ayata   S.-D. , Lindsay   D. J. , Karp-Boss   L. , Stemmann   L.   2022 . Machine learning for the study of plankton and marine snow from images . Annual Review of Marine Science , 14 : 277 – 301 .

James   G. , Witten   D. , Hastie   T. , Tibshirani   R.   2013 . An Introduction to Statistical Learning . Springer , New York, NY .

Jamet   C. , Thiria   S. , Moulin   C. , Crepon   M.   2005 . Use of a neurovariational inversion for retrieving oceanic and atmospheric constituents from ocean color imagery: a feasibility study . Journal of Atmospheric and Oceanic Technology , 22 : 460 – 475 .

Jarvis   S. , DiMarzio   N. , Morrissey   R. , Moretti   D.   2008 . A novel multi-class support vector machine classifier for automated classification of beaked whales and other small odontocetes . Canadian Acoustics , 36 : 34 – 40 .

Jordan   M. I. , Mitchell   T. M.   2015 . Machine learning: trends, perspectives, and prospects . Science , 349 : 255 – 260 .

Katija   K. , Orenstein   E. , Schlining   B. , Lundsten   L. , Barnard   K. , Sainz   G. , Boulais   O  et al.    2022 . FathomNet: a global image database for enabling artificial intelligence in the ocean . Scientific Reports , 12 : 15914 .

Kerr   T. , Clark   J. R. , Fileman   E. S. , Widdicombe   C. E. , Pugeault   N.   2020 . Collaborative deep learning models to handle class imbalance in flowcam plankton imagery . IEEE Access , 8 : 170013 – 170032 .

Kikaki   K. , Kakogeorgiou   I. , Mikeli   P. , Raitsos   D. E. , Karantzalos   K.   2022 . MARIDA: a benchmark for marine debris detection from sentinel-2 remote sensing data . PLoS One , 17 : e0262247 .

Kiranyaz   S. , Gabbouj   M. , Pulkkinen   J. , Ince   T. , Meissner   K.   2010 . Network of evolutionary binary classifiers for classification and retrieval in macroinvertebrate databases . IEEE International Conference on Image Processing , pp. 2257 – 2260 .. IEEE .

Kiranyaz   S. , Ince   T. , Pulkkinen   J. , Gabbouj   M. , Ärje   J. , Kärkkäinen   S. , Tirronen   V  et al.    2011 . Classification and retrieval on macroinvertebrate image databases . Computers in Biology and Medicine , 41 : 463 – 472 .

Knudby   A.   2010 . New approaches to modelling fish–habitat relationships . Ecological Modelling , 221 : 503 – 511 .

Knudby   A. , LeDrew   E. , Brenning   A.   2010 . Predictive mapping of reef fish species richness, diversity and biomass in Zanzibar using IKONOS imagery and machine-learning techniques . Remote Sensing of Environment , 114 : 1230 – 1241 .

Korneliussen   R. J. (Ed.) 2018 . Acoustic target classification . ICES Cooperative Research Report , 344, 104 pp.

Kouw   W. M. , Loog   M.   2019 . A review of domain adaptation without target labels . IEEE Transactions on Pattern Analysis and Machine Intelligence , 43 : 766 – 785 .

Kowarski   K. A. , Moors-Murphy   H.   2021 . A review of big data analysis methods for baleen whale passive acoustic monitoring . Marine Mammal Science , 37 ( 2 ), pp. 652 – 673 .

Krizhevsky   A. , Sutskever   I. , Hinton   G. E.   2012 . Imagenet classification with deep convolutional neural networks. In . Advances in Neural Information Processing Systems , pp. 1097 – 1105 .. Ed. by Pereira   F. , Burges   C. J. C. , Bottou   L. , Weinerger   K. Q. . Neural Information Processing Systems Foundation, Inc. (NeurIPS) , Lake Tahoe, NV .

Kroodsma   D. A. , Mayorga   J. , Hochberg   T. , Miller   N. A. , Boerder   K. , Ferretti   F. , Wilson   A  et al.    2018 . Tracking the global footprint of fisheries . Science , 359 : 904 – 908 .

Kuhn   M. , Wickham   H.   2020 . Tidymodels: a collection of packages for modeling and machine learning using tidyverse principles . https://www.tidymodels.org  (accessed 29 September 2022) .

Kyathanahally   S. P. , Hardeman   T. , Reyes   M. , Merz   E. , Bulas   T. , Brun   P. , Pomati   F  et al.    2022 . Ensembles of data-efficient vision transformers as a new paradigm for automated classification in ecology . Scientific Reports , 12 : 18590 .

Landschützer   P. , Gruber   N. , Haumann   F. A. , Rödenbeck   C. , Bakker   D. C. E. , van Heuven   S. , Hoppema   M  et al.    2015 . The reinvigoration of the Southern Ocean carbon sink . Science , 349 : 1221 – 1224 .

Langenkämper   D. , van Kevelaer   R. , Purser   A. , Nattkemper   T. W.   2020 . Gear-induced concept drift in marine images and its effect on deep learning classification . Frontiers in Marine Science , 7 : 506 .

Langenkämper   D. , Zurowietz   M. , Schoening   T. , Nattkemper   T. W.   2017 . BIIGLE 2.0—browsing and annotating large marine image collections . Frontiers in Marine Science , 4 : 83.

Laurila-Pant   M. , Mäntyniemi   S. , Venesjärvi   R. , Lehikoinen   A.   2019 . Incorporating stakeholders’ values into environmental decision support: a Bayesian belief network approach . Science of The Total Environment , 697 : 134026 .

Le Guyader   D. , Ray   C. , Brosset   D.   2016 . Defining fishing grounds variability with automatic identification system (AIS) . 2nd International Workshop on Maritime Flows and Networks (WIMAKS’16) , p. 96 .

LeCun   Y. , Bengio   Y. , Hinton   G.   2015 . Deep learning . Nature , 521 : 436 – 444 .

Lecun   Y. , Bottou   L. , Bengio   Y. , Haffner   P.   1998 . Gradient-based learning applied to document recognition . Proceedings of the IEEE , 86 : 2278 – 2324 .

Lee   H. , Park   M. , Kim   J.   2016 . Plankton classification on imbalanced large scale database via convolutional neural networks with transfer learning . 2016 IEEE International Conference on Image Processing (ICIP) , pp. 3713 – 3717 .

Lehikoinen   A. , Olsson   J. , Bergström   L. , Bergström   U. , Bryhn   A. , Fredriksson   R. , Uusitalo   L.   2019 . Evaluating complex relationships between ecological indicators and environmental factors in the Baltic Sea: a machine learning approach . Ecological Indicators , 101 : 117 – 125 .

Lekunberri   X. , Ruiz   J. , Quincoces   I. , Dornaika   F. , Arganda-Carreras   I. , Fernandes   J. A.   2022 . Identification and measurement of tropical tuna species in purse seiner catches using computer vision and deep learning . Ecological Informatics , 67 : 101495 .

Li   Y. , Guo   J. , Guo   X. , Hu   Z. , Tian   Y.   2021 . Plankton detection with adversarial learning and a densely connected deep learning model for class imbalanced distribution . Journal of Marine Science and Engineering , 9 : 636 .

Lieshout   C. , Oeveren   K. , Emmerik   T. , Postma   E.   2020 . Automated river plastic monitoring using deep learning and cameras . Earth and Space Science , 7 : e2019EA000960 .

Lima-Mendez   G. , Faust   K. , Henry   N. , Decelle   J. , Colin   S. , Carcillo   F. , Chaffron   S  et al.    2015 . Determinants of community structure in the global plankton interactome . Science , 348 : 1262073 .

Lin   T.-H. , Yang   H.-T. , Huang   J.-M. , Yao   C.-J. , Lien   Y.-S. , Wang   P.-J. , Hu   F.-Y.   2019 . Evaluating changes in the marine soundscape of an offshore wind farm via the machine learning-based source separation . In   2019 IEEE Underwater Technology (UT) , pp. 1 – 6 .

Liu   H. , Li   Q. , Bai   Y. , Yang   C. , Wang   J. , Zhou   Q. , Hu   S  et al.    2021 . Improving satellite retrieval of oceanic particulate organic carbon concentrations using machine learning methods . Remote Sensing of Environment , 256 : 112316 .

Liu   Y. , Wang   S.   2021 . A quantitative detection algorithm based on improved faster R-CNN for marine benthos . Ecological Informatics , 61 : 101228 .

Liu   Y. , Weisberg   R. H.   2011 . A review of self-organizing map applications in meteorology and oceanography . In   Self Organizing Maps—Applications and Novel Algorithm Design . Ed. by   Mwasiagi   J. I . InTech .

Liu   Y. , Xu   J. , Tao   Y. , Fang   T. , Du   W. , Ye   A.   2020 . Rapid and accurate identification of marine microbes with single-cell Raman spectroscopy . The Analyst , 145 : 3297 – 3305 .

Longhurst   A. , Sathyendranath   S. , Platt   T. , Caverhill   C.   1995 . An estimate of global primary production in the ocean from satellite radiometer data . Journal of Plankton Research , 17 : 1245 – 1271 .

Lopez-Vazquez   V. , Lopez-Guede   J. M. , Marini   S. , Fanelli   E. , Johnsen   E. , Aguzzi   J.   2020 . Video image enhancement and machine learning pipeline for underwater animal detection and classification at cabled observatories . Sensors , 20 : 726 .

Lucas   T. C. D.   2020 . A translucent box: interpretable machine learning in ecology . Ecological Monographs , 90 : e01422 .

Lumini   A. , Nanni   L. , Maguolo   G.   2020 . Deep learning for plankton and coral classification . Applied Computing and Informatics , 19 : 265 – 283 .

Madricardo   F. , Ghezzo   M. , Nesto   N. , Mc Kiver   W. J. , Faussone   G. C. , Fiorin   R. , Riccato   F  et al.    2020 . How to deal with seafloor marine litter: an overview of the state-of-the-art and future perspectives . Frontiers in Marine Science , 7 : 505134 .

Mahé   F. , Rognes   T. , Quince   C. , de Vargas   C. , Dunthorn   M.   2015 . Swarm v2: highly-scalable and high-resolution amplicon clustering . PeerJ , 3 : e1420 .

Malde   K. , Handegard   N. O. , Eikvil   L. , Salberg   A.-B.   2020 . Machine intelligence and the data-driven future of marine science . ICES Journal of Marine Science , 77 : 1274 – 1285 .

Maldonado   A. D. , Uusitalo   L. , Tucker   A. , Blenckner   T. , Aguilera   P. A. , Salmerón   A.   2019 . Prediction of a complex system with few data: evaluation of the effect of model structure and amount of data with dynamic bayesian network models . Environmental Modelling and Software , 118 : 281 – 297 .

Malfante   M. , Mohammed   O. , Gervaise   C. , Mura   M. D. , Mars   J. I.   2018 . Use of deep features for the automatic classification of fish sounds . 2018 OCEANS—MTS/IEEE Kobe Techno-Oceans (OTO) , IEEE , pp. 1 – 5 .

Marques   T. P. , Rezvanifar   A. , Cote   M. , Albu   A. B. , Ersahin   K. , Mudge   T. , Gauthier   S.   2021 . Detecting marine species in echograms via traditional, hybrid, and deep learning frameworks . 2020 25th International Conference on Pattern Recognition (ICPR) , IEEE , pp. 5928 – 5935 .

Marzuki   M. I. , Gaspar   P. , Garello   R. , Kerbaol   V. , Fablet   R.   2018 . Fishing gear identification from vessel-monitoring-system-based fishing vessel trajectories . IEEE Journal of Oceanic Engineering , 43 : 689 – 699 .

Mattei   F. , Franceschini   S. , Scardi   M.   2018 . A depth-resolved artificial neural network model of marine phytoplankton primary production . Ecological Modelling , 382 : 51 – 62 .

Mayot   N. , D'Ortenzio   F. , Ribera d'Alcalà   M. , Lavigne   H. , Claustre   H.   2016 . Interannual variability of the Mediterranean trophic regimes from ocean color satellites . Biogeosciences , 13 : 1901 – 1917 .

Mitchell   T. M.   1997 . Machine Learning . McGraw-Hill , New York, NY .

Mitchell   T. M.   1999 . Machine learning and data mining . Communications of the ACM , 42 : 30 – 36 .

Moreno-Torres   J. G. , Raeder   T. , Alaiz-Rodríguez   R. , Chawla   N. V. , Herrera   F.   2012 . A unifying view on dataset shift in classification . Pattern Recognition , 45 : 521 – 530 .

Muñoz   M. , Reul   A. , Vargas-Yáñez   M. , Plaza   F. , Bautista   B. , García-Martínez   M. C. , Moya   F  et al.    2017 . Fertilization and connectivity in the Garrucha Canyon (SE-Spain) implications for marine spatial planning . Marine Environmental Research , 126 : 45 – 68 .

Murali   A. , Bhargava   A. , Wright   E. S.   2018 . IDTAXA: a novel approach for accurate taxonomic classification of microbiome sequences . Microbiome , 6 : 140 .

Niu   H. , Reeves   E. , Gerstoft   P.   2017 . Source localization in an ocean waveguide using supervised machine learning . The Journal of the Acoustical Society of America , 142 : 1176 – 1188 .

NOAA . 2014 . Report on the occurrence and health effects of anthropogenic debris ingested by marine organisms . Silver Spring , MD . 19 pp .

NOAA . 2016 . Report on modeling oceanic transport of floating marine debris . Silver Spring , MD . 21 pp .

Olden   J. D. , Lawler   J. J. , Poff   N. L.   2008 . Machine learning methods without tears: a primer for ecologists . The Quarterly Review of Biology , 83 : 171 – 193 .

Orenstein   E. C. , Beijbom   O.   2017 . Transfer learning and deep feature extraction for planktonic image data sets . In   2017 IEEE Winter Conference on Applications of Computer Vision (WACV) , pp. 1082 – 1088 .

Orenstein   E. C. , Kenitz   K. M. , Roberts   P. L. D. , Franks   P. J. S. , Jaffe   J. S. , Barton   A. D.   2020 . Semi- and fully supervised quantification techniques to improve population estimates from machine classifiers . Limnology and Oceanography: Methods , 18 : 739 – 753 .

Ozanich   E. , Thode   A. , Gerstoft   P. , Freeman   L. A. , Freeman   S.   2021 . Deep embedded clustering of coral reef bioacoustics . The Journal of the Acoustical Society of America , 149 : 2587 – 2601 .

Özel Duygan   B. D. , Hadadi   N. , Babu   A. F. , Seyfried   M. , van der Meer   J. R.   2020 . Rapid detection of microbiota cell type diversity using machine-learned classification of flow cytometry data . Communications Biology , 3 : 379 .

Paszke   A. , Gross   S. , Massa   F. , Lerer   A. , Bradbury   J. , Chanan   G. , Killeen   T  et al.    2019 . Pytorch: an imperative style, high-performance deep learning library . Advances in Neural Information Processing Systems , 32 .

Pedregosa   F. , Varoquaux   G. , Gramfort   A. , Michel   V. , Thirion   B. , Grisel   O. , Blondel   M  et al.    2011 . Scikit-learn: machine learning in Python . Journal of Machine Learning Research , 12 : 2825 – 2830 .

Peña   M.   2018 . Robust clustering methodology for multi-frequency acoustic data: a review of standardization, initialization and cluster geometry . Fisheries Research , 200 : 49 – 60 .

Phillips   L. R. , Carroll   G. , Jonsen   I. , Harcourt   R. , Roughan   M.   2020 . A water mass classification approach to tracking variability in the east Australian current . Frontiers in Marine Science , 7 : 365 .

Phillips   S. J. , Anderson   R. P. , Schapire   R. E.   2006 . Maximum entropy modeling of species geographic distributions . Ecological Modelling , 190 : 231 – 259 .

Phillips   S. J. , Dudík   M.   2008 . Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation . Ecography , 31 : 161 – 175 .

Picheral   M. , Colin   S. , Irisson   J.-O . 2017 . EcoTaxa, a tool for the taxonomic classification of images . http://ecotaxa.obs-vlfr.fr  (accessed 29 September 2022) .

Pichler   M. , Boreux   V. , Klein   A. , Schleuning   M. , Hartig   F.   2020 . Machine learning algorithms to infer trait-matching and predict species interactions in ecological networks . Methods in Ecology and Evolution , 11 : 281 – 293 .

Piechaud   N. , Hunt   C. , Culverhouse   P. F. , Foster   N. L. , Howell   K. L.   2019 . Automated identification of benthic epifauna with computer vision . Marine Ecology Progress Series , 615 : 15 – 30 .

Pınarbaşı   K. , Galparsoro   I. , Borja   Á. , Stelzenmüller   V. , Ehler   C. N. , Gimpel   A.   2017 . Decision support tools in marine spatial planning: present applications, gaps and future perspectives . Marine Policy , 83 : 83 – 91 .

Pınarbaşı   K. , Galparsoro   I. , Depellegrin   D. , Bald   J. , Pérez-Morán   G. , Borja   Á.   2019 . A modelling approach for offshore wind farm feasibility with respect to ecosystem-based marine spatial planning . Science of the Total Environment , 667 : 306 – 317 .

Plonus   R.-M. , Conradt   J. , Harmer   A. , Janssen   S. , Floeter   J.   2021 . Automatic plankton image classification-Can capsules and filters help cope with data set shift? . Limnology and Oceanography: Methods , 19 : 176 – 195 .

Politikos   D. V. , Adamopoulou   A. , Petasis   G. , Galgani   F.   2023 . Using artificial intelligence to support marine macrolitter research: a content analysis and an online database . Ocean and Coastal Management , 233 : 106466 .

Politikos   D. V. , Fakiris   E. , Davvetas   A. , Klampanos   I. A. , Papatheodorou   G.   2021 . Automatic detection of seafloor marine litter using towed camera images and deep learning . Marine Pollution Bulletin , 164 : 111974 .

Ponsero   A. J. , Hurwitz   B. L.   2019 . The promises and pitfalls of machine learning for detecting viruses in aquatic metagenomes . Frontiers in Microbiology , 10 : 806.

Porskamp   P. , Rattray   A. , Young   M. , Ierodiaconou   D.   2018 . Multiscale and hierarchical classification for benthic habitat mapping . Geosciences , 8 : 119 .

Preston   F. W.   1948 . The commonness, and rarity, of species . Ecology , 29 : 254 – 283 .

Quang   D. , Xie   X.   2016 . DanQ: a hybrid convolutional and recurrent deep neural network for quantifying the function of DNA sequences . Nucleic Acids Research , 44 : e107 – e107 .

Quast   C. , Pruesse   E. , Yilmaz   P. , Gerken   J. , Schweer   T. , Yarza   P. , Peplies   J  et al.    2013 . The SILVA ribosomal RNA gene database project: improved data processing and web-based tools . Nucleic Acids Research , 41 : D590 – D596 .

Quinn   T. P. , Erb   I. , Gloor   G. , Notredame   C. , Richardson   M. F. , Crowley   T. M.   2019 . A field guide for the compositional analysis of any-omics data . GigaScience , 8 : giz107 .

Rajwa   B. , Venkatapathi   M. , Ragheb   K. , Banada   P. P. , Hirleman   E. D. , Lary   T. , Robinson   J. P.   2008 . Automated classification of bacterial particles in flow by multiangle scatter measurement and support vector machine classifier . Cytometry Part A , 73A : 369 – 379 .

Reichstein   M. , Camps-Valls   G. , Stevens   B. , Jung   M. , Denzler   J. , Carvalhais   N. , Prabhat . 2019 . Deep learning and process understanding for data-driven Earth system science . Nature , 566 : 195 .

Reiss   H. , Cunze   S. , König   K. , Neumann   H. , Kröncke   I.   2011 . Species distribution modelling of marine benthos: a North Sea case study . Marine Ecology Progress Series , 442 : 71 – 86 .

Reygondeau   G. , Guidi   L. , Beaugrand   G. , Henson   S. A. , Koubbi   P. , MacKenzie   B. R. , Sutton   T. T  et al.    2018 . Global biogeochemical provinces of the mesopelagic zone . Journal of Biogeography , 45 : 500 – 514 .

Rezvanifar   A. , Marques   T. P. , Cote   M. , Albu   A. B. , Slonimer   A. , Tolhurst   T. , Ersahin   K. , et al.   2019 . A deep learning-based framework for the detection of schools of herring in echograms . arXiv:1910.08215 [cs, eess, stat] . http://arxiv.org/abs/1910.08215  ( accessed 6 August 2021 ).

Richards   C. , Cooke   R. S. , Bowler   D. E. , Boerder   K. , Bates   A. E.   2021 . Bycatch mitigation could prevent strong changes in the ecological strategies of seabird communities across the globe . https://doi.org/10.1101/2021.05.24.445481 .

Robards   M. D. , Silber   G. K. , Adams   J. D. , Arroyo   J. , Lorenzini   D. , Schwehr   K. , Amos   J.   2016 . Conservation science and policy applications of the marine vessel Automatic Identification System (AIS)—a review . Bulletin of Marine Science , 92 : 75 – 103 .

Robinson   K. L. , Luo   J. Y. , Sponaugle   S. , Guigand   C. , Cowen   R. K.   2017 . A tale of two crowds: public engagement in plankton classification . Frontiers in Marine Science , 4 : 82 .

Roch   M. A. , Klinck   H. , Baumann-Pickering   S. , Mellinger   D. K. , Qui   S. , Soldevilla   M. S. , Hildebrand   J. A.   2011 . Classification of echolocation clicks from odontocetes in the Southern California Bight . The Journal of the Acoustical Society of America , 129 : 467 – 475 .

Roch   M. A. , Lindeneau   S. , Aurora   G. S. , Frasier   K. E. , Hildebrand   J. A. , Glotin   H. , Baumann-Pickering   S.   2021 . Using context to train time-domain echolocation click detectors . The Journal of the Acoustical Society of America , 149 : 3301 – 3310 .

Rognes   T. , Flouri   T. , Nichols   B. , Quince   C. , Mahé   F.   2016 . VSEARCH: a versatile open source tool for metagenomics . PeerJ , 4 : e2584 .

Rose   G. A. , Leggett   W. C.   1988 . Hydroacoustic signal classification of fish schools by species . Canadian Journal of Fisheries and Aquatic Sciences , 45 : 597 – 604 .

Roshan   S. , DeVries   T.   2017 . Efficient dissolved organic carbon production and export in the oligotrophic ocean . Nature Communications , 8 : 2036 .

Rowell   T. J. , Demer   D. A. , Aburto-Oropeza   O. , Cota-Nieto   J. J. , Hyde   J. R. , Erisman   B. E.   2017 . Estimating fish abundance at spawning aggregations from courtship sound levels . Scientific Reports , 7 : 3340 .

Rubbens   P. , Props   R.   2021 . Computational analysis of microbial flow cytometry data . mSystems , 6 : e00895 – 20 .,.

Rubbens   P. , Props   R. , Boon   N. , Waegeman   W.   2017 . Flow cytometric single-cell identification of populations in synthetic bacterial communities . PLoS One , 12 : e0169754 .

Rubbens   P. , Props   R. , Kerckhof   F.-M. , Boon   N. , Waegeman   W.   2021 . PhenoGMM: Gaussian mixture modeling of cytometry data quantifies changes in microbial community structure . mSphere , 6 : e00530 – 20 .,.

Rudin   C.   2019 . Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead . Nature Machine Intelligence , 1 : 206 – 215 .

Russo   T. , Franceschini   S. , D'Andrea   L. , Scardi   M. , Parisi   A. , Cataudella   S.   2019 . Predicting fishing footprint of trawlers from environmental and fleet data: an application of artificial neural networks . Frontiers in Marine Science , 6 : 670.

Russo   T. , Parisi   A. , Garofalo   G. , Gristina   M. , Cataudella   S. , Fiorentino   F.   2014 . SMART: a spatially explicit bio-economic model for assessing and managing demersal fisheries, with an application to italian trawlers in the strait of sicily . PLoS One , 9 : e86222 .

Russo   T. , Parisi   A. , Prorgi   M. , Boccoli   F. , Cignini   I. , Tordoni   M. , Cataudella   S.   2011 . When behaviour reveals activity: assigning fishing effort to métiers based on VMS data using artificial neural networks . Fisheries Research , 111 : 53 – 64 .

Samuel   A. L.   1959 . Some studies in machine learning using the game of checkers . IBM Journal of Research and Development , 3 : 210 – 229 .

Sander   E. L. , Wootton   J. T. , Allesina   S.   2017 . Ecological network inference from long-term presence–absence data . Scientific Reports , 7 : 7154 .

Santora.   2012 . Spatial ecology of krill, micronekton and top predators in the central California Current: implications for defining ecologically important areas . Progress in Oceanography , 106 : 154 – 174 .

Santos-Domínguez   D. , Torres-Guijarro   S. , Cardenal-López   A. , Pena-Gimenez   A.   2016 . ShipsEar: an underwater vessel noise database . Applied Acoustics , 113 : 64 – 69 .

Sauzède   R. , Claustre   H. , Jamet   C. , Uitz   J. , Ras   J. , Mignot   A. , D'Ortenzio   F.   2015a . Retrieving the vertical distribution of chlorophyll a concentration and phytoplankton community composition from in situ fluorescence profiles: a method based on a neural network with potential for global-scale applications . Journal of Geophysical Research: Oceans , 120 : 451 – 470 .

Sauzède   R. , Lavigne   H. , Claustre   H. , Uitz   J. , Schmechtig   C. , d’Ortenzio   F. , Guinet   C.  et al.    2015b . Vertical distribution of chlorophyll a concentration and phytoplankton community composition from in situ fluorescence profiles: a first database for the global ocean . Earth System Science Data , 7 : 261 – 273 .

Sauzède   R. , Claustre   H. , Uitz   J. , Jamet   C. , Dall'Olmo   G. , D'Ortenzio   F. , Gentili   B  et al.    2016 . A neural network-based method for merging ocean color and Argo data to extend surface bio-optical properties to depth: retrieval of the particulate backscattering coefficient . Journal of Geophysical Research: Oceans , 121 : 2552 – 2571 .

Sauzède   R. , Bittig   H. C. , Claustre   H. , Pasqueron de Fommervault   O. , Gattuso   J.-P. , Legendre   L. , Johnson   K. S.   2017 . Estimates of water-column nutrient concentrations and carbonate system parameters in the global ocean: a novel approach based on neural networks . Frontiers in Marine Science , 4 : 128.

Schmarje   L. , Santarossa   M. , Schröder   S.-M. , Zelenka   C. , Kiko   R. , Stracke   J. , Volkmann   N  et al.    2022 . A data-centric approach for improving ambiguous labels with combined semi-supervised classification and clustering . Computer Vision—ECCV 2022: 17th European Conference, Tel Aviv, Israel, October 23–27, 2022 , Part VIII , pp. 363 – 380 .. Berlin, Heidelberg . https://doi.org/10.1007/978-3-031-20074-8_21   ( accessed 29 November 2022 ).

Schroeder   S.-M. , Kiko   R. , Koch   R.   2020 . MorphoCluster: efficient annotation of plankton images by clustering . Sensors , 20 : 3060 . Basel .

Schröter   H. , Nöth   E. , Maier   A. , Cheng   R. , Barth   V. , Bergler   C.   2019 . Segmentation, classification, and visualization of orca calls using deep learning . 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pp. 8231 – 8235 .

Sgier   L. , Freimann   R. , Zupanic   A. , Kroll   A.   2016 . Flow cytometry combined with viSNE for the analysis of microbial biofilms and detection of microplastics . Nature Communications , 7 : 11587 .

Shafait   F. , Mian   A. , Shortis   M. , Ghanem   B. , Culverhouse   P. F. , Edgington   D. , Cline   D  et al.    2016 . Fish identification from videos captured in uncontrolled underwater environments . ICES Journal of Marine Science: Journal du Conseil , 73 : 2737 – 2746 .

Shang   Y. , Li   J.   2018 . Study on echo features and classification methods of fish species . 2018 10th International Conference on Wireless Communications and Signal Processing (WCSP) , pp. 1 – 6 .

Shao   H. , Kiyomoto   S. , Kawauchi   Y. , Kadota   T. , Nakagawa   M. , Yoshimura   T. , Yamada   H  et al.    2021 . Classification of various algae canopy, algae turf, and barren seafloor types using a scientific echosounder and machine learning analysis . Estuarine, Coastal and Shelf Science , 255 : 107362 .

Sharma   G. , Umapathy   K. , Krishnan   S.   2020 . Trends in audio signal feature extraction methods . Applied Acoustics , 158 : 107020 .

Shin   D.   2021 . The effects of explainability and causability on perception, trust, and acceptance: implications for explainable AI . International Journal of Human-Computer Studies , 146 : 102551 .

Smith   J. A. , Tommasi   D. , Welch   H. , Hazen   E. L. , Sweeney   J. , Brodie   S. , Muhling   B  et al.    2021 . Comparing dynamic and static time-area closures for bycatch mitigation: a management strategy evaluation of a swordfish fishery . Frontiers in Marine Science , 8 : 630607.

Smoliński   S. , Radtke   K.   2017 . Spatial prediction of demersal fish diversity in the Baltic Sea: comparison of machine learning and regression-based techniques . ICES Journal of Marine Science , 74 : 102 – 111 .

Solsona-Berga   A. , Frasier   K. E. , Baumann-Pickering   S. , Wiggins   S. M. , Hildebrand   J. A.   2020 . DetEdit: a graphical user interface for annotating and editing events detected in long-term acoustic monitoring data . PLoS Computational Biology , 16 : e1007598 .

Sonnewald   M. , Dutkiewicz   S. , Hill   C. , Forget   G.   2020 . Elucidating ecological complexity: unsupervised learning determines global marine eco-provinces . Science Advances , 6 : eaay4740 .

Soriano   M. , Marcos   S. , Saloma   C. , Quibilan   M. , Alino   P.   2001 . Image classification of coral reef components from underwater color video . In   MTS/IEEE Oceans 2001 . An Ocean Odyssey. Conference Proceedings (IEEE Cat. No.01CH37295) , pp. 1008 – 1013 .  vol. 2 .

Sosik   H. M. , Olson   R. J.   2007 . Automated taxonomic classification of phytoplankton sampled with imaging-in-flow cytometry . Limnology and Oceanography: Methods , 5 : 204 – 216 .

Sosik   H. M. , Peacock   E. E. , Brownlee   E. F.   2015 . WHOI plankton, annotated plankton images—Data set for developing and evaluating classification methods . http://hdl.handle.net/1912/7341  (accessed 10 April 2021) .

Soykan   C. U. , Eguchi   T. , Kohin   S. , Dewar   H.   2014 . Prediction of fishing effort distributions using boosted regression trees . Ecological Applications , 24 : 71 – 83 .

Spampinato   C. , Giordano   D. , Di Salvo   R. , Chen-Burger   Y.-H. J. , Fisher   R. B. , Nadarajan   G.   2010 . Automatic fish classification for underwater species behavior understanding . Proceedings of the First ACM International Workshop on Analysis and Retrieval of Tracked Events and Motion in Imagery Streams , pp. 45 – 50 .. New York, NY . http://doi.acm.org/10.1145/1877868.1877881   (accessed 7 August 2019) .

Stephens   D. , Diesing   M.   2014 . A comparison of supervised classification methods for the prediction of substrate type using multibeam acoustic and legacy grain-size data . PLoS One , 9 : e93950 .

Stewart   W. K. , Jiang   M. , Marra   M.   1994 . A neural network approach to classification of sidescan sonar imagery from a midocean ridge area . IEEE Journal of Oceanic Engineering , 19 : 214 – 224 .

Stock   A. , Subramaniam   A.   2020 . Accuracy of empirical satellite algorithms for mapping phytoplankton diagnostic pigments in the open ocean: a supervised learning perspective . Frontiers in Marine Science , 7 : 599.

Stock   B. C. , Ward   E. J. , Eguchi   T. , Jannot   J. E. , Thorson   J. T. , Feist   B. E. , Semmens   B. X.   2020 . Comparing predictions of fisheries bycatch using multiple spatiotemporal species distribution model frameworks . Canadian Journal of Fisheries and Aquatic Sciences , 77 : 146 – 163 .

Storbeck   F. , Daan   B.   2001 . Fish species recognition using computer vision and a neural network . Fisheries Research , 51 : 11 – 15 .

Stowell   D.   2022 . Computational bioacoustics with deep learning: a review and roadmap . PeerJ , 10 : e13152 . [ CrossRef ]

Suikkanen   S. , Uusitalo   L. , Lehtinen   S. , Lehtiniemi   M. , Kauppila   P. , Mäkinen   K. , Kuosa   H.   2021 . Diazotrophic cyanobacteria in planktonic food webs . Food Webs , 28 : e00202 .

Taconet   M. , Kroodsma   D. , Fernandes   J. A. , Food and Agriculture Organization of the United Nations, Global Fishing Watch, AZTI-Tecnalia, and Seychelles Fishing Authority . 2019 . Global atlas of AIS-based fishing activity: challenges and opportunities . 395 pp. www.fao.org/3/ca7012en/ca7012en.pdf  (accessed 29 September 2022) .

Tang   W. , Li   Z. , Cassar   N.   2019 . Machine learning estimates of global marine nitrogen fixation . Journal of Geophysical Research: Biogeosciences , 124 : 717 – 730 .

Tang   X. , Stewart   W. K. , Huang   H. , Gallager   S. M. , Davis   C. S. , Vincent   L. , Marra   M.   1998 . Automatic plankton image recognition . Artificial Intelligence Review , 12 : 177 – 199 .

Tanhua   T. , Pouliquen   S. , Hausman   J. , O'Brien   K. , Bricher   P. , de Bruin   T. , Buck   J. J. H. , et al.    2019 . Ocean FAIR data services . Frontiers in Marine Science , 6 : 440 .

Thomas   M. K. , Fontana   S. , Reyes   M. , Pomati   F.   2018 . Quantifying cell densities and biovolumes of phytoplankton communities and functional groups using scanning flow cytometry, machine learning and unsupervised clustering . PLoS One , 13 : e0196225 .

Thomas   M. , Martin   B. , Kowarski   K. , Gaudet   B. , Matwin   S.   2020 . Marine mammal species classification using convolutional neural networks and a novel acoustic representation . In   Machine Learning and Knowledge Discovery in Databases , pp. 290 – 305 .. Ed. by   Brefeld   U. , Fromont   E. , Hotho   A. , Knobbe   A. , Maathuis   M. , Robardet   C . Springer International Publishing , Cham .

Thoya   P. , Maina   J. , Möllmann   C. , Schiele   K. S.   2021 . AIS and VMS ensemble can address data gaps on fisheries for marine spatial planning . Sustainability , 13 : 3769 .

Trifonova   N. , Kenny   A. , Maxwell   D. , Duplisea   D. , Fernandes   J. , Tucker   A.   2015 . Spatio-temporal Bayesian network models with latent variables for revealing trophic dynamics and functional networks in fisheries ecology . Ecological Informatics , 30 : 142 – 158 .

Trifonova   N. , Maxwell   D. , Pinnegar   J. , Kenny   A. , Tucker   A.   2017 . Predicting ecosystem responses to changes in fisheries catch, temperature, and primary productivity with a dynamic Bayesian network model . ICES Journal of Marine Science , 74 : 1334 – 1343 .

Tseng   C.-H. , Kuo   Y.-F.   2020 . Detecting and counting harvested fish and identifying fish types in electronic monitoring system videos using deep convolutional neural networks . ICES Journal of Marine Science , 77 : 1367 – 1378 .

Uusitalo   L. , Fernandes   J. A. , Bachiller   E. , Tasala   S. , Lehtiniemi   M.   2016 . Semi-automated classification method addressing marine strategy framework directive (MSFD) zooplankton indicators . Ecological Indicators , 71 : 398 – 405 .

Uusitalo   L. , Tomczak   M. T. , Müller-Karulis   B. , Putnis   I. , Trifonova   N. , Tucker   A.   2018 . Hidden variables in a dynamic Bayesian network identify ecosystem level change . Ecological Informatics , 45 : 9 – 15 .

Vacher   C. , Tamaddoni-Nezhad   A. , Kamenova   S. , Peyrard   N. , Moalic   Y. , Sabbadin   R. , Schwaller   L  et al.    2016 . Learning ecological networks from next-generation sequencing data . In   Advances in Ecological Research , 54 : 1 – 39 .

Vestbo   S. , Obst   M. , Quevedo Fernandez   F. J. , Intanai   I. , Funch   P.   2018 . Present and potential future distributions of Asian horseshoe crabs determine areas for conservation . Frontiers in Marine Science , 5 :164.

Villon   S. , Mouillot   D. , Chaumont   M. , Darling   E. S. , Subsol   G. , Claverie   T. , Villéger   S.   2018 . A deep learning method for accurate and fast identification of coral reef fishes in underwater images . Ecological Informatics , 48 : 238 – 244 .

Wang   Q. , Garrity   G. M. , Tiedje   J. M. , Cole   J. R.   2007 . Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy . Applied and Environmental Microbiology , 73 : 5261 – 5267 .

Watanabe   J. , Shao   Y. , Miura   N.   2019 . Underwater and airborne monitoring of marine ecosystems and debris . Journal of Applied Remote Sensing , 13 : 1 .

Watling   J. I. , Brandt   L. A. , Bucklin   D. N. , Fujisaki   I. , Mazzotti   F. J. , Romañach   S. S. , Speroterra   C.   2015 . Performance metrics and variance partitioning reveal sources of uncertainty in species distribution models . Ecological Modelling , 309–310 : 48 – 59 .

Weber   L. M. , Saelens   W. , Cannoodt   R. , Soneson   C. , Hapfelmeier   A. , Gardner   P. P. , Boulesteix   A.-L  et al.    2019 . Essential guidelines for computational method benchmarking . Genome Biology , 20 : 125 .

Weilgart   L. , Whitehead   H.   1997 . Group-specific dialects and geographical variation in coda repertoire in South Pacific sperm whales . Behavioral Ecology and Sociobiology , 40 : 277 – 285 .

Welch   H. , Hazen   E. L. , Bograd   S. J. , Jacox   M. G. , Brodie   S. , Robinson   D. , Scales   K. L  et al.    2019 . Practical considerations for operationalizing dynamic management tools . Journal of Applied Ecology , 56 : 459 – 469 .

Welch   H. , McHenry   J.   2018 . Planning for dynamic process: an assemblage-level surrogate strategy for species seasonal movement pathways . Aquatic Conservation: Marine and Freshwater Ecosystems , 28 : 337 – 350 .

Welch   H. , Pressey   R. L. , Heron   S. F. , Ceccarelli   D. M. , Hobday   A. J.   2016 . Regimes of chlorophyll-a in the Coral Sea: implications for evaluating adequacy of marine protected areas . Ecography , 39 : 289 – 304 .

Welch   H. , Pressey   R. L. , Reside   A. E.   2018 . Using temporally explicit habitat suitability models to assess threats to mobile species and evaluate the effectiveness of marine protected areas . Journal for Nature Conservation , 41 : 106 – 115 .

White   T. D. , Ong   T. , Ferretti   F. , Block   B. A. , McCauley   D. J. , Micheli   F. , De Leo   G. A.   2020 . Tracking the response of industrial fishing fleets to large marine protected areas in the Pacific Ocean . Conservation Biology , 34 : 1571 – 1578 .

Wick   R. R. , Judd   L. M. , Holt   K. E.   2019 . Performance of neural network basecalling tools for Oxford nanopore sequencing . Genome Biology , 20 : 129 .

Williams   S. , Friedman   A.   2018 . SQUIDLE+ . http://squidle.acfr.usyd.edu.au  (accessed 29 September 2022) .

Wirbel   J. , Zych   K. , Essex   M. , Karcher   N. , Kartal   E. , Salazar   G. , Bork   P  et al.    2021 . Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT machine learning toolbox . Genome Biology , 22 : 93 .

Yoon   B.-J.   2009 . Hidden markov models and their applications in biological sequence analysis . Current Genomics , 10 : 402 – 415 .

Yuan   Q. , Shen   H. , Li   T. , Li   Z. , Li   S. , Jiang   Y. , Xu   H  et al.    2020 . Deep learning in environmental remote sensing: achievements and challenges . Remote Sensing of Environment , 241 : 111716 .

Zaugg   S. , van der Schaar   M. , Houégnigan   L. , Gervaise   C. , André   M.   2010 . Real-time acoustic classification of sperm whale clicks and shipping impulses from deep-sea observatories . Applied Acoustics , 71 : 1011 – 1019 .

Zhang   C. , Selch   D. , Xie   Z. , Roberts   C. , Cooper   H. , Chen   G.   2013 . Object-based benthic habitat mapping in the Florida keys from hyperspectral imagery . Estuarine, Coastal and Shelf Science , 134 : 88 – 97 .

Zhao   Q. , Costello   M. J.   2019 . Summer and winter ecosystems of the world ocean photic zone . Ecological Research , 34 ( 4 ): 457 – 471 .

Zion   B. , Alchanatis   V. , Ostrovsky   V. , Barki   A. , Karplus   I.   2007 . Real-time underwater sorting of edible fish species . Computers and Electronics in Agriculture , 56 : 34 – 45 .

Supplementary data

Email alerts, citing articles via.

• Contact ICES
• Recommend to your Library

Affiliations

• Online ISSN 1095-9289
• Print ISSN 1054-3139
• Copyright © 2024 ICES/CIEM
• About Oxford Academic
• Publish journals with us
• University press partners
• What we publish
• New features  
• Open access
• Institutional account management
• Rights and permissions
• Get help with access
• Accessibility
• Advertising
• Media enquiries
• Oxford University Press
• Oxford Languages
• University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide

• Copyright © 2024 Oxford University Press
• Cookie settings
• Cookie policy
• Privacy policy
• Legal notice

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This PDF is available to Subscribers Only

For full access to this pdf, sign in to an existing account, or purchase an annual subscription.

share this!

March 21, 2024

This article has been reviewed according to Science X's editorial process and policies . Editors have highlighted the following attributes while ensuring the content's credibility:

fact-checked

peer-reviewed publication

trusted source

Scientists detail research to assess the viability and risks of marine cloud brightening

by Peter Genzer, Brookhaven National Laboratory

As the levels of greenhouse gases in the atmosphere continue to increase and climate change impacts become more costly, the scientific community is redoubling efforts to investigate the potential risks and benefits of artificially shading Earth's surface to slow global warming.

Marine cloud brightening (MCB) is one of two primary solar radiation modification methods being proposed to offset the worst effects of global warming while decarbonization advances. MCB proposals involve the injection of salt spray into shallow marine clouds to brighten them, increasing their reflection of sunlight and reducing the amount of heat absorbed by the water below.

A group of 31 leading atmospheric scientists are now offering a consensus physical science research roadmap to build the knowledge base needed to evaluate the viability of MCB approaches. Their roadmap is described in a new paper published in the journal Science Advances .

"Interest in MCB is growing, but policymakers currently don't have the information they need to reach decisions about if and when MCB should be deployed," said lead author Graham Feingold, a researcher with NOAA's Chemical Sciences Laboratory.

"The question is whether we can design a MCB research program using our current modeling and observational tools to establish the feasibility of this approach on a global scale, and if not, what needs to be done to position ourselves to do so."

Artificially shading the planet would do nothing to reduce the driver of climate change, human-caused greenhouse gas emissions, said co-author Lynn Russell, a climate scientist at the Scripps Institution of Oceanography at the University of California San Diego.

"The recent acceleration of impacts from global warming means that we need to consider non-ideal backup plans just to buy us enough time to reduce greenhouse gas emissions and existing burdens," Russell said. "A research plan is essential before we can consider adopting MCB, and we need to simultaneously address the physical science questions and the human dimensions."

Current MCB proposals rely on saltwater spray, which would mimic plumes of sulfur-rich emissions from ship stacks or volcanoes, to increase the aerosol concentration in the lower marine atmosphere. Ideally, droplets in the saltwater spray evaporate to produce fine particles that are carried up to the cloud layer by turbulent and convective air motions.

If MCB techniques could consistently influence clouds to reflect more sunlight back to space than similar clouds with a lower droplet concentration, then it has the potential to be an effective solar radiation modification technique, at least at the local scale, scientists say. This in turn could produce some cooling at a local scale.

The study proposes a substantial and targeted program of MCB research that includes laboratory studies, field experiments, and cloud modeling. As a result, new laboratory facilities are needed to address gaps in understanding aerosol and cloud microphysical processes, as few existing labs are capable of addressing these processes.

Long-running field experiments using a point source at an ocean-based location where the conditions are favorable, along with new observations and new modeling are needed to test salt-particle spraying technology. This would allow scientists to determine the degree to which sea spray emitted near the surface would reach the cloud base in a variety of conditions.

Researchers can take advantage of existing analogs to cloud-seeding experiments, such as natural volcanic emissions, biomass burning, exhaust plumes from individual ships or designated shipping lanes, urban point sources, and urban plumes.

In practical terms, researchers need to develop sufficient confidence that appropriately sized particles can be generated and delivered to the clouds, and once there, act to form cloud droplets that efficiently scatter sunlight. They would need to show that clouds could be brightened consistently and over a large enough area to meaningfully cool the ocean below—and that trying to manipulate clouds would not cause clouds to thin, or droplets to rain out, which might allow for increased heating.

Scientists would further need to show that the brightening of the clouds would be measurable to demonstrate it would work as intended at globally relevant scales, or in sensitive regional ecosystems, such as coral reefs.

Clouds are not all created equally—some are more susceptible to aerosol injections than others. A cloud that is already bright, with a high drop concentration, is much more difficult to brighten than a wispy cloud with a low drop concentration. How a cloud responds to attempted manipulation is subtly dependent on the weather and background aerosol conditions.

Complicating matters, the optimal particle size and amount is likely dependent on cloud properties that can change as they drift through the air. This explains the high variability in ship-track occurrence, Feingold said.

"We would have to get the right-sized particles into receptive clouds at the right times of day and seasons, and over large enough areas to shade large areas of ocean," said Feingold. "It's a major challenge."

"To the extent that we can identify optimal brightening conditions, a targeted approach to MCB, rather than routine spraying under all conditions, might have a higher probability of success," Feingold said. "It might also reduce the risk of regional circulation responses that change temperature and rainfall in ways that benefit some and leave others vulnerable."

More generally, Feingold re-emphasizes that MCB would not replace decarbonization and would not alleviate ocean acidification. "To reduce global temperatures, our highest priority should be to remove carbon dioxide from the atmosphere. MCB might help to alleviate the worst impacts of climate change ."

Journal information: Science Advances

Provided by Brookhaven National Laboratory

Explore further

Feedback to editors

Heat to blame for space pebble demise

9 hours ago

New cost-effective method can detect low concentrations of pharmaceutical waste and contaminants in water

10 hours ago

Team proposes using AI to reconstruct particle paths leading to new physics

A new way to quantify climate change impacts: 'Outdoor days'

Higher temperatures mean higher food and other prices. A new study links climate shocks to inflation

Satellite data assimilation improves forecasts of severe weather

Scientists create novel technique to form human artificial chromosomes

Shakespeare's sister: Digital archives reveal hidden insights into world-famous playwright's unknown sibling

Research reveals new starting points for the rapid and targeted development of future drugs

High speed protein movies to aid drug design

Relevant physicsforums posts, unlocking the secrets of prof. verschure's rosetta stones, iceland warming up again - quakes swarming.

11 hours ago

Higher Chance to get Lightning Strike by Large Power Consumption?

Mar 20, 2024

A very puzzling rock or a pallasite / mesmosiderite or a nothing burger

Mar 16, 2024

Earth's earliest forest discovered in SW England

Mar 8, 2024

La Cumbre volcano eruption, Fernandina, Galapagos Islands

Mar 4, 2024

More from Earth Sciences

Related Stories

Could 'marine cloud brightening' reduce coral bleaching on the Great Barrier Reef?

Oct 17, 2023

New cloud model could help with climate research

Feb 21, 2024

Aerosol particles cool the climate less than we thought

Jan 28, 2021

Research finds marine bacteria, atmospheric rivers can contribute to formation of ice clouds

Dec 8, 2023

Climate predictions require increasingly accurate information on atmospheric particles

Nov 10, 2023

Cloud study demystifies impact of aerosols

Aug 1, 2022

Recommended for you

Study suggests millions are at risk using high arsenic water for cooking

13 hours ago

AI could help predict floods where traditional methods struggle

15 hours ago

Research reveals global wildfire risk trends in wildland–urban interface areas

17 hours ago

Let us know if there is a problem with our content

Use this form if you have come across a typo, inaccuracy or would like to send an edit request for the content on this page. For general inquiries, please use our contact form . For general feedback, use the public comments section below (please adhere to guidelines ).

Please select the most appropriate category to facilitate processing of your request

Thank you for taking time to provide your feedback to the editors.

Your feedback is important to us. However, we do not guarantee individual replies due to the high volume of messages.

E-mail the story

Your email address is used only to let the recipient know who sent the email. Neither your address nor the recipient's address will be used for any other purpose. The information you enter will appear in your e-mail message and is not retained by Phys.org in any form.

Newsletter sign up

Get weekly and/or daily updates delivered to your inbox. You can unsubscribe at any time and we'll never share your details to third parties.

More information Privacy policy

Donate and enjoy an ad-free experience

We keep our content available to everyone. Consider supporting Science X's mission by getting a premium account.

E-mail newsletter

Help | Advanced Search

Computer Science > Computer Vision and Pattern Recognition

Title: end-to-end underwater video enhancement: dataset and model.

Abstract: Underwater video enhancement (UVE) aims to improve the visibility and frame quality of underwater videos, which has significant implications for marine research and exploration. However, existing methods primarily focus on developing image enhancement algorithms to enhance each frame independently. There is a lack of supervised datasets and models specifically tailored for UVE tasks. To fill this gap, we construct the Synthetic Underwater Video Enhancement (SUVE) dataset, comprising 840 diverse underwater-style videos paired with ground-truth reference videos. Based on this dataset, we train a novel underwater video enhancement model, UVENet, which utilizes inter-frame relationships to achieve better enhancement performance. Through extensive experiments on both synthetic and real underwater videos, we demonstrate the effectiveness of our approach. This study represents the first comprehensive exploration of UVE to our knowledge. The code is available at https://anonymous.4open.science/r/UVENet.

Submission history

Access paper:.

• Download PDF
• HTML (experimental)
• Other Formats

References & Citations

• Google Scholar
• Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

Scientists detail research to assess viability and risks of marine cloud brightening

• March 20, 2024

As the levels of greenhouse gases in the atmosphere continue to increase and climate change impacts become more costly, the scientific community is redoubling efforts to investigate the potential risks and benefits of artificially shading Earth’s surface to slow global warming. Marine cloud brightening (MCB) is one of two primary solar radiation modification methods being proposed to offset the worst effects of global warming while decarbonization advances. MCB proposals involve the injection of salt spray into shallow marine clouds to brighten them, increasing their reflection of sunlight and reducing the amount of heat absorbed by the water below.

A group of 31 leading atmospheric scientists have now offered a consensus physical science research roadmap to build the knowledge base needed to evaluate the viability of MCB approaches. Their roadmap is described in a new paper published in the journal Science Advances .

“Interest in MCB is growing, but policymakers currently don’t have the information they need to reach decisions about if and when MCB should be deployed,” said lead author Graham Feingold, a researcher with NOAA’s Chemical Sciences Laboratory. “The question is whether we can design a MCB research program using our current modeling and observational tools to establish the feasibility of this approach on a global scale, and if not, what needs to be done to position ourselves to do so.” 

Artificially shading the planet would do nothing to reduce the driver of climate change, human-caused greenhouse gas emissions, said co-author Lynn Russell, a climate scientist at the Scripps Institution of Oceanography at the University of California San Diego. “The recent acceleration of impacts from global warming means that we need to consider non-ideal backup plans just to buy us enough time to reduce greenhouse gas emissions and existing burdens,” Russell said. “A research plan is essential before we can consider adopting MCB, and we need to simultaneously address the physical science questions and the human dimensions.”

Current MCB proposals rely on saltwater spray, which would mimic plumes of sulfur-rich emissions from ship stacks or volcanoes, to increase the aerosol concentration in the lower marine atmosphere. Ideally, droplets in the saltwater spray evaporate to produce fine particles that are carried up to the cloud layer by turbulent and convective air motions. If MCB techniques could consistently influence clouds to reflect more sunlight back to space than similar clouds with a lower droplet concentration, then it has the potential to be an effective solar radiation modification technique, at least at the local scale, scientists say. This in turn could produce some cooling at a local scale.

The study proposes a substantial and targeted program of MCB research that includes laboratory studies, field experiments, and cloud modeling. As a result, new laboratory facilities are needed to address gaps in understanding aerosol and cloud microphysical processes, as few existing labs are capable of addressing these processes.

Long-running field experiments using a point source at an ocean-based location where the conditions are favorable, along with new observations and new modeling are needed to test salt-particle spraying technology. This would allow scientists to determine the degree to which sea spray emitted near the surface would reach the cloud base in a variety of conditions.

Researchers can take advantage of existing analogs to cloud-seeding experiments, such as natural volcanic emissions, biomass burning, exhaust plumes from  individual ships or designated shipping lanes, urban point sources, and urban plumes. 

In practical terms, researchers need to develop sufficient confidence that appropriately sized particles can be generated and delivered to the clouds, and once there, act to form cloud droplets that efficiently scatter sunlight. They would need to show that clouds could be brightened consistently and over a large enough area to meaningfully cool the ocean below – and that trying to manipulate clouds would not cause clouds to thin, or droplets to rain out, which might allow for increased heating. Scientists would further need to show that the brightening of the clouds would be measurable to demonstrate it would work as intended at globally relevant scales, or in sensitive regional ecosystems, such as coral reefs. 

Clouds are not all created equally—some are more susceptible to aerosol injections than others. A cloud that is already bright, with a high drop concentration, is much more difficult to brighten than a wispy cloud with a low drop concentration. How a cloud responds to attempted manipulation is subtly dependent on the weather and background aerosol conditions. Complicating matters, the optimal particle size and amount is likely dependent on cloud properties that can change as they drift through the air. This explains the high variability in ship-track occurrence, Feingold said. 

“We would have to get the right-sized particles into receptive clouds at the right times of day and seasons, and over large-enough areas to shade large areas of ocean,” said Feingold. “It’s a major challenge.”

“To the extent that we can identify optimal brightening conditions, a targeted approach to MCB, rather than routine spraying under all conditions, might have a higher probability of success,” Feingold said. “It might also reduce the risk of regional circulation responses that change temperature and rainfall in ways that benefit some and leave others vulnerable.”

More generally, Feingold cautioned that MCB would not replace decarbonization and would not alleviate ocean acidification. “To reduce global temperatures, our highest priority should be to remove carbon dioxide from the atmosphere. MCB might help to alleviate the worst impacts of climate change.”

For more information, contact Theo Stein, NOAA Communications: [email protected] .

How social science helps us combat climate change

Could drying the stratosphere help cool the planet?

50 years of getting ENSO predictions *mostly* correct

Great Lakes ice coverage reaches historic low

Popup call to action.

A prompt with more information on your call to action.

• Share full article

Advertisement

Supported by

Scientists Discover 100 New Marine Species in New Zealand

The findings, from the largely uncharted waters of Bounty Trough, show that “we’ve got a long way to go in terms of understanding where life is found in the ocean,” a researcher said.

By Rebecca Carballo

A team of 21 scientists set off on an expedition in the largely uncharted waters of Bounty Trough off the coast of the South Island of New Zealand in February hoping to find a trove of new species.

The expedition paid off, they said on Sunday, with the discovery of 100 new species, a number that was likely to grow, said Alex Rogers, a marine biologist who was a leader of the expedition.

“I expect that number to increase as we work through more and more of the samples,” Dr. Rogers said. “I think that number is going to be in the hundreds instead of just 100.”

Dozens of mollusks, three fish, a shrimp and a cephalopod that is a type of predatory mollusk were among the new species found in the expedition, which was led by Ocean Census, a nonprofit dedicated to the global discovery of ocean life, the National Institute of Water and Atmospheric Research in New Zealand, and the Museum of New Zealand Te Papa Tongarewa.

One creature that caused a “lot of head-scratching” is a star-shaped animal, about a centimeter across, but researchers have not managed to identify it, Dr. Rogers said. They believe it may possibly be a coral.

Two million-plus species are estimated to live in the oceans, but only 10 percent of ocean life is known. It is vital to learn more about the aquatic life because marine ecosystems carry out functions that support life on Earth, such as creating food for billions, storing carbon and regulating climate, Dr. Rogers said.

“We’re dealing with a situation where we know marine life is in decline,” he said. “In order to try to manage human activities to prevent this continuing decline, we need to understand the distribution of marine life better than we currently do.”

Ocean Census was founded last year by the Nippon Foundation, a Japanese philanthropic organization, and the U.K.-based ocean exploration foundation Nekton. When it began its work, Ocean Census set a goal of finding at least 100,000 new marine species in a decade.

The group is focused on exploring some of the most under-sampled bodies of water.

In the February expedition, researchers first mapped the area with an imaging system and video cameras to check that it would be safe for their equipment and to ensure that there were no vulnerable animal communities that potentially could be harmed.

Then, they deployed what is known as the Brenke sled , a sampling device that has two nets, one close to the seabed, and the other a meter above it. As it drags along the floor, it churns up animals living close to the sea floor. To find larger animals, the researchers used other methods, such as baited nets.

Trawling the depths at 4,800 meters — or roughly the equivalent to Mont Blanc, the highest peak in the Alps — researchers collected 1,791 samples.

Given its depth, Bounty Trough is not of great interest to fisheries and therefore is poorly sampled, Dr. Rogers said. Geologists have surveyed this area but biologists have not.

Worldwide, about 240,000 marine species have been discovered and named to date but only 2,200 species are discovered each year on average, according to Ocean Census.

In many bodies of water there is still a lot that scientists have to learn, Dr. Rogers said.

“It’s probably the equivalent of a space mission,” he said. “We’re still in early days, but the number of species that we found in the Bounty Trough really indicates to us that we’ve got a long way to go in terms of understanding where life is found in the ocean.”

Rebecca Carballo is a reporter based in New York. More about Rebecca Carballo

Explore the Animal Kingdom

A selection of quirky, intriguing and surprising discoveries about animal life..

Aside from chimps and humans, researchers have found clear evidence of menopause in only five species — all of them whales. A new study looks at the possible causes for it .

Scientists never imagined that the blind cave salamanders called olms willingly left their caves. Then, they discovered several at aboveground springs in northern Italy .

According to a common narrative that male mammals tend to be larger than female ones. A new study paints a more complex picture .

Daddy longlegs, the group of splendidly leggy arachnids also known as harvestmen, have been thought to have just two eyes. New research has uncovered four more vestigial ones .

The means by which some whales sing underwater has long been a mystery. A contraption that forced air through the larynxes of three carcasses puts forth an explanation .

Here’s how a male elephant seal, not usually possessed with a paternal instinct, prevented a younger animal from drowning in an unlikely act of altruism .

Join Our Whale-y Big Celebration!

We’re celebrating whales all month long!

Dive in with us during the month of March as we highlight the largest marine mammals: whales! Learn fun facts about these gentle giants, what conservation efforts are helping to protect them, ways you can be a whale champion and more.

• Research Library

Explore Our Innovative Medical Techniques and Vital Scientific Research

Featured research projects, leptospirosis, domoic acid toxicosis, the marine mammal center’s experts are leading contributors to global knowledge about marine mammal health, including developing innovative medical techniques and conducting vital research that has revealed alarming ocean health concerns., all publications.

Use the form below to search for all publications

Recent Publications

Research Paper

New Urban Habitat for Endangered Humpback Whales: San Francisco Bay

Fish feeding and rapid foraging behavior switching by gray whales (eschrichtius robustus) in california, novel presentation of coccidioidomycosis (valley fever) in a southern sea otter, marine mammal and marine bird surveys during the windfloat pacific offshore wind project, sexual behavior and anatomy in porpoises, sociosexual behavior of nocturnally foraging dusky and spinner dolphins.

• Reference Manager
• Simple TEXT file

People also looked at

Editorial article, editorial: sustainable development goal 14 - life below water: towards a sustainable ocean.

• 1 Oceans and Atmosphere, Commonwealth Scientific and Industrial Research Organisation (CSIRO), Perth, WA, Australia
• 2 Centre for Blue Governance, University of Portsmouth, Portsmouth, United Kingdom
• 3 UWA Oceans Institute and School of Biological Sciences, The University of Western Australia, Crawley, WA, Australia
• 4 AZTI, Marine Research, Basque Research and Technology Alliance (BRTA), Herrera Kaia, Portualdea z/g, Pasaia, Spain
• 5 Faculty of Marine Sciences, King Abdulaziz University, Jeddah, Saudi Arabia
• 6 Ocean Conservancy, Washington, DC, United States
• 7 School of Environmental Sciences, University of East Anglia, Norwich, United Kingdom
• 8 Aarhus University, Aarhus, Denmark
• 9 National Autonomous University of Mexico, México City, Mexico
• 10 Georgia Institute of Technology, Atlanta, GA, United States
• 11 NIVA Denmark Water Research, Copenhagen, Denmark
• 12 Department of Oceanography, Federal University of Pernambuco, Recife, Brazil
• 13 Centre for Environment, Fisheries and Aquaculture Science (CEFAS), Lowestoft, United Kingdom
• 14 Department of Applied Economics, University of Santiago de Compostela, Santiago de Compostela, Spain
• 15 Department of Oceanography, National Institute of Oceanography and Applied Geophysics (Istituto Nazionale di Oceanografia e di Geofisica Sperimentale) OGS, Trieste, Italy
• 16 Italian Institute for Environmental Protection and Research (Istituto Superiore per la Protezione e la Ricerca Ambientale) - ISPRA, Ozzano dell'Emilia, Italy

Editorial on the Research Topic Sustainable Development Goal 14 - Life Below Water: Towards a Sustainable Ocean

United Nations (UN) Sustainable Development Goal (SDG) 14 – Life Below Water – is arguably one of the most challenging of the 17 goals ( United Nations, 2016 ) due to the immense scale of the Ocean (almost three-quarters of the planet's surface) and the direct links to many other SDGs. For example, No Poverty (SDG 1), Zero Hunger (SDG2) and Good Health and Well-Being (SDG 3) all rely on sustainable Life Below Water (SDG 14). In turn, Climate Action (SDG 13) is needed to achieve SDG 14, and the Ocean is essential in achieving SDG 13. There is much that we still do not know; indeed, the Ocean represents more than 99% of the space where organisms can live, yet more than 80% of the Ocean remains unexplored, especially the deep-sea.

The launch of the UN Decade of Ocean Science for Sustainable Development (2021–2030) aims at catalyzing a global focus to advance SDG 14 ( Borja et al., 2020a ). This will enhance the co-design of knowledge and actions for transformative ocean solutions, to address the challenges of a growing human population and climate change. Human pressures on the Ocean are important – 37% of the human population live in the coast from small villages to megacities exceeding 10 million people (e.g., New York, Shanghai, Lagos) and use the Ocean for a huge range of inputs, outputs and services, including amenity, food, transport, cooling water and waste disposal, as well as traditional and cultural uses. Many of these ecosystem services are undervalued, being conservatively estimated at $12.6 Trillion annually more than 20 years ago ( Costanza et al., 1997 ). This is without considering two of the most severely undervalued services provided by the Ocean, as heat and carbon sinks, that have buffered many of the negative impacts of climate change. Many anthropogenic activities are leaving significant, direct and measurable global footprints in the Ocean with high profile examples including fishing 1 , 2 , 3 , , shipping lanes ( Liu et al., 2019 ; Pirotta et al., 2019 ), dredging 4 , plastic pollution ( Hardesty et al., 2017 ; Barrett et al., 2020 ), noise pollution ( Di Franco et al., 2020 ; Chahouri et al., 2021 ; Duarte et al., 2021 ), and changes in Ocean chemistry 5 .

Human populations rely directly on the Ocean for food and other commercial activities, but a growing body of research has identified our dependency on the Ocean for health and well-being ( Borja et al., 2020b ). Other ecosystem services provided by the Ocean are also yet to be properly considered. These include the cultural and spiritual services provided by the Ocean ( Brown and Hausner, 2017 ; de Juan et al., 2021 ), which have developed over millennia of human relationships with the Ocean and represent knowledge and connections that extend beyond monetary value. Aiming to integrate this knowledge in scientific endeavours, many indigenous peoples are bringing their traditional science and knowledge to partner with western science ( Mazzocchi, 2006 ) and provide a more in-depth and long-term understanding of the Ocean, especially in coastal areas ( Mustonen et al., 2021 ).

While the challenges are clear and sometimes seem overwhelming, approaches and solutions are being actively developed and tested; several of these are explored in this Research Topic.

With more than three billion people who rely on fish for at least 20% of their daily protein, and more than 120 million directly employed in the fishing and aquaculture sectors 6 , sustainable fishing ( Penca ; Fiorentino and Vitale ; Jaiteh et al. ) and aquaculture ( Azra et al. ) were a natural focus of several papers. This included a call for reducing effort in mixed species fisheries, and therefore fishing mortality, to take into account the differing and lower productivity of some species and the risk to their sustainability ( Newman et al., 2018 ), and adopt a quota system based on “pretty good yield” ( Hilborn, 2010 ).

Others emphasized the need for better conservation planning and coordination ( Katsanevakis et al. ; Ceccarelli et al. ; Herrera et al. ) as well as integration of their cultural and spiritual values into wider society ( Baker et al. ). This includes the need to improve spatial management, providing specific approaches to minimize human impacts and risks to charismatic megafauna. This management approach could be applied to whale watching activities, to support sustainable non-extractive human activities in the Ocean ( Almunia et al. ). The article by Adewumi et al. , dealing with the Guinea Current Large Marine Ecosystem shared among Benin, Nigeria, and Cameroon, highlighted the challenges of international ocean governance, a result of political characteristics, the relics of colonialism, and increasing ocean use and pressure on marine ecosystems and services. The administrative and political arrangements differ significantly among countries, complicating transnational collaboration. The review of these arrangements revealed varying levels of convergence at international, regional and national levels, and could be a model to assist regional fishery management organizations to support positive steps toward ocean sustainability ( Juan-Jordá et al., 2018 ).

Future risks to the Ocean ( Garcia-Soto et al. ), including those imposed by climate change ( Green et al. ), and the tools ( Mariani et al. ), approaches (e.g., Endrédi et al. ; Hsu et al. ), and ways to monitor this complex system ( Jones et al. ), including biodiversity ( Herrera et al. ), highlighted the extraordinary and diverse values of the Ocean and challenges ( Figure 1 ). Embracing modern technologies ( Almunia et al. ; Green et al. ), including the Internet of Things ( Mariani et al. ), could also promote and support a harmonization of ocean monitoring among all nations, and support international initiatives and cooperation 7 , including platforms to involve the wider community 8 .

Figure 1 . Word cloud generated from the key words from the SDG 14 papers contributed toward this Frontiers in Marine Science Research Topic [Generated through WordArt.com - Word Cloud Art Creator]. Some key words were truncated to simplify the generation of the word cloud.

The social dimension ( Haward and Haas ) will also be critical as a way of valuing and engaging with direct and indirect stakeholders of the Ocean and in developing better policies for governance ( Paredes-Coral et al. ; Polejack ; Adewumi et al. ; Kirkfeldt and Frazão Santos ; Archana and Baker ; Rohmana et al. ). This is especially true at the land-sea interface ( Singh et al. ) where human populations concentrate and the risks from a changing climate are directly evident, with projected sea level rise ( Nicholls and Cazenave, 2010 ; Hooijer and Vernimmen, 2021 ), and more frequent and intense storms ( Pugatch, 2019 ; Chen et al., 2020 ). It is also true for the deep ocean ( Howell et al. ), which remains largely unexplored. The socio-ecological connections described in this Research Topic of Frontiers in Marine Science provide frameworks and hope for a sustainable future for the coasts and ocean.

While this Frontiers in Marine Science Research Topic does not represent all initiatives underway globally to address SDG 14, it provides a glimpse of some of the diverse approaches and intellectual capital invested in ocean sustainability. While the goal focuses on Life Below Water, these approaches directly support many other SDGs, which arguably cannot be achieved without a healthy and sustainable ocean ( Mustonen et al., 2021 ).

We hope that other initiatives currently underway will assist in not only highlighting the links between SDG 14 and other SDGs but also provide a way for synergies among disparate knowledge domains to support transdisciplinary and multi-sectoral approaches for good policy development. As examples, we note the significant initiatives around the globe in areas of blue carbon and an equitable “blue economy.” Blue carbon projects not only protect and restore seagrass, mangrove, salt marsh, and macrophytes, but also support the associated biodiversity and human livelihoods that depend on these critical habitat-forming species. “Working with nature approaches” including in the restoration of corals, seagrasses, seaweeds, and mangroves are underway around the globe, with new methods being developed and tested [e.g., genetic techniques to identify more heat tolerant species of coral ( Buerger et al., 2020 ) and other marine habitat building species ( Alsuwaiyan et al., 2021 )].

The efforts in these areas will be underpinned by new methods of accounting—such as blue carbon, biodiversity, ecosystem services and a framework of ocean accounting which is currently being developed 9 . This approach embraces environmental, social and cultural accounting, in addition to economic accounting, to better assess and value entire marine areas and ecosystems and integrate a wide range of SDGs. Our hope is that this will support and enable clearer and better decisions by ocean and coastal management agencies. These decisions should be based on a number of decision support tools, including: (i) management strategy evaluation approaches, (ii) scenario testing including assessing a range of alternative approaches, and (iii) potentially creating digital twins to test and explore management decisions before ocean activities commence.

We look forward to making the difficult possible and contributing to a vibrant, thriving future throughout the UN Decade of Ocean Science for Sustainable Development and the UN Decade of Restoration ( Waltham et al., 2020 ) based on some of the cutting-edge approaches detailed in this Research Topic of Frontiers in Marine Science .

Author Contributions

All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.

Conflict 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.

Publisher's Note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

1. ^ https://globalfishingwatch.org/ .

2. ^ http://www.seaaroundus.org/ .

3. ^ https://www.minderoo.org/global-fishing-index/ .

4. ^ https://wamsi.org.au/wp-content/uploads/bsk-pdf-manager/2019/10/Dredging-Science-Synthesis-Report-A-Synthesis-of-Research-2012-2018-April-2019.pdf .

5. ^ https://www.science.org.au/curious/earth-environment/ocean-acidification .

6. ^ https://www.fao.org/in-action/eaf-nansen/news-events/detail-events/en/c/1413988/ .

7. ^ https://www.geoaquawatch.org/ .

8. ^ https://research.csiro.au/eyeonwater/ .

9. ^ https://www.oceanaccounts.org/ .

Alsuwaiyan, N. A., Vranken, S., Filbee-Dexter, K., Cambridge, M., Coleman, M. A., and Wernberg, T. (2021). Genotypic variation in response to extreme events may facilitate kelp adaptation under future climates. Mar. Ecol. Progr. Ser . 672, 111–121. doi: 10.3354/meps13802

CrossRef Full Text | Google Scholar

Barrett, J., Chase, Z., Zhang, J., Holl, M. M. B., Willis, K., Williams, A., et al. (2020). Microplastic pollution in deep-sea sediments from the Great Australian Bight. Front. Mar. Sci . 7:576170. doi: 10.3389/fmars.2020.576170

Borja, A., Andersen, J. H., Arvanitidis, C. D., Basset, A., Buhl-Mortensen, L., Carvalho, S., et al. (2020a). Past and future grand challenges in marine ecosystem ecology. Front. Mar. Sci . 7:362. doi: 10.3389/fmars.2020.00362

Borja, A., White, M. P., Berdalet, E., Bock, N., Eatock, C., Kristensen, P., et al. (2020b). Moving toward an agenda on ocean health and human health in Europe. Front. Mar. Sci . 7:37. doi: 10.3389/fmars.2020.00037

Brown, G., and Hausner, V. H. (2017). An empirical analysis of cultural ecosystem values in coastal landscapes. Ocean Coast. Manag . 142, 49–60. doi: 10.1016/j.ocecoaman.03.019

Buerger, P., Alvarez-Roa, C., Coppin, C. W., Pearce, S. L., Chakravarti, L. J., Oakeshott, J. G., et al. (2020). Heat-evolved microalgal symbionts increase coral bleaching tolerance. Sci. Adv . 6:eaba2498. doi: 10.1126/sciadv.aba2498

PubMed Abstract | CrossRef Full Text | Google Scholar

Chahouri, A., Elouahmani, N., and Ouchene, H. (2021). Recent progress in marine noise pollution: a thorough review. Chemosphere 2021:132983. doi: 10.1016/j.chemosphere.2021.132983

Chen, J., Wang, Z., Tam, C. Y., Lau, N. C., Lau, D. S., Lau, D., et al. (2020). Impacts of climate change on tropical cyclones and induced storm surges in the Pearl River Delta region using pseudo-global-warming method. Sci. Rep . 10:1965. doi: 10.1038/s41598-020-58824-8

Costanza, R., d'Arge, R., de Groot, R., Farber, S., Grasso, M., Hannon, B., et al. (1997). The value of the world's ecosystem services and natural capital. Nature 387, 253–260. doi: 10.1038/387253a0

de Juan, S., Ospina-Álvarez, A., Villasante, S., and Ruiz-Frau, A. (2021). A graph theory approach to assess nature's contribution to people at a global scale. Sci. Rep. 11, 9118. doi: 10.1038/s41598-021-88745-z

Di Franco, C., Faverney, R., Rossi, F., Sabourault, C., Spennato, G., Verrando, P., et al. (2020). Effects of marine noise pollution on Mediterranean fishes and invertebrates: a review. Mar. Poll. Bullet . 159:111450. doi: 10.1016/j.marpolbul.2020.111450

Duarte, C. M., Chapuis, L., Collin, S. P., Costa, D. P., Devassy, R. P., Eguiluz, V. M., et al. (2021). The soundscape of the Anthropocene ocean. Science . 371:eaba4658. doi: 10.1126/science.aba4658

Hardesty, B. D., Lawson, T. J., van der Velde, T., Lansdell, M., and Wilcox, C. (2017). Estimating quantities and sources of marine debris at a continental scale. Front. Ecol. Environ . 15, 18–25. doi: 10.1002/fee.1447

Hilborn, R.. (2010). Pretty Good Yield and exploited fishes. Mar. Pol . 34, 193–196. doi: 10.1016/j.marpol.04, 013.

Hooijer, A., and Vernimmen, R. (2021). Global LiDAR land elevation data reveal greatest sea-level rise vulnerability in the tropics. Nat. Commun . 12:3592. doi: 10.1038/s41467-021-23810-9

Juan-Jordá, M. J., Murua, H., Arrizabalaga, H., Dulvy, N. K., and Restrepo, V. (2018). Report card on ecosystem-based fisheries management in tuna regional fisheries management organizations. Fish Fisheries 19, 321–339. doi: 10.1111/faf.12256

Liu, H., Meng, Z. H., Lv, Z. F., Wang, X. T., Deng, Y., Liu, Y., et al. (2019). Emissions and health impacts from global shipping embodied in US–China bilateral trade. Nat. Sustainabil . 2, 1027–1033. doi: 10.1038/s41893-019-0414-z

Mazzocchi, F.. (2006). Western science and traditional knowledge. Despite their variations, different forms of knowledge can learn from each other. EMBO Rep . 7, 463–466. doi: 10.1038/sj.embor.7400693

Mustonen, T., Maxwell, K. H., Mustonen, K., Jones, R., Pedersen, H., Nuunoq, J. G., et al. (2021). Who is the ocean? Preface to the future seas 2030 special issue. Rev. Fish Biol. Fisheries 21:9655. doi: 10.1007/s11160-021-09655-x

Newman, S. J., Brown, J. I., Fairclough, D. V., Wise, B. S., Bellchambers, L. M., Molony, B. W., et al. (2018). A risk assessment and prioritisation approach to the selection of indicator species for the assessment of multi-species, multi-gear, multi-sector fishery resources. Mar. Pol . 88, 11–22. doi: 10.1016/j.marpol.10, 028.

Nicholls, R. J., and Cazenave, A. (2010). Sea-level rise and its impact on coastal zones. Science 328, 1517–1520. doi: 10.1126/science.1185782

Pirotta, V., Grech, A., Jonsen, I. D., Laurance, W. F., and Harcourt, R. G. (2019). Consequences of global shipping traffic for marine giants. Front. Ecol. Environ . 17, 39–47. doi: 10.1002/fee.1987

Pugatch, T.. (2019). Tropical storms and mortality under climate change. World Dev . 117, 172–182. doi: 10.1016/j.worlddev.2019.01.009

United Nations (2016). Report of the Inter-Agency and Expert Group on Sustainable Development Goal Indicators. (E/CN.3/2016/2/Rev.1) . New York, NY: United Nations Economic and Social Council, 49.

Google Scholar

Waltham, N. J., Elliott, M., Lee, S. Y., Lovelock, C., Duarte, C. M., Buelow, C., et al. (2020). UN decade on ecosystem restoration 2021–2030—what chance for success in restoring coastal ecosystems? Front. Mar. Sci. 7:71. doi: 10.3389/fmars.2020.00071

Keywords: Sustainable Development Goal 14 - Life Below Water, SDGs, ocean, livelihoods, United Nation (UN)

Citation: Molony BW, Ford AT, Sequeira AMM, Borja A, Zivian AM, Robinson C, Lønborg C, Escobar-Briones EG, Di Lorenzo E, Andersen JH, Müller MN, Devlin MJ, Failler P, Villasante S, Libralato S and Fortibuoni T (2022) Editorial: Sustainable Development Goal 14 - Life Below Water: Towards a Sustainable Ocean. Front. Mar. Sci. 8:829610. doi: 10.3389/fmars.2021.829610

Received: 06 December 2021; Accepted: 29 December 2021; Published: 24 February 2022.

Copyright © 2022 Molony, Ford, Sequeira, Borja, Zivian, Robinson, Lønborg, Escobar-Briones, Di Lorenzo, Andersen, Müller, Devlin, Failler, Villasante, Libralato and Fortibuoni. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Brett W. Molony, brett.molony@csiro.au

This article is part of the Research Topic

Sustainable Development Goal 14 - Life Below Water: Towards a Sustainable Ocean

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Published: 07 July 2022

A global horizon scan of issues impacting marine and coastal biodiversity conservation

• James E. Herbert-Read   ORCID: orcid.org/0000-0003-0243-4518 1   na1 ,
• Ann Thornton   ORCID: orcid.org/0000-0002-7448-8497 2   na1 ,
• Diva J. Amon   ORCID: orcid.org/0000-0003-3044-107X 3 , 4 ,
• Silvana N. R. Birchenough   ORCID: orcid.org/0000-0001-5321-8108 5 ,
• Isabelle M. Côté   ORCID: orcid.org/0000-0001-5368-4061 6 ,
• Maria P. Dias   ORCID: orcid.org/0000-0002-7281-4391 7 , 8 ,
• Brendan J. Godley 9 ,
• Sally A. Keith   ORCID: orcid.org/0000-0002-9634-2763 10 ,
• Emma McKinley   ORCID: orcid.org/0000-0002-8250-2842 11 ,
• Lloyd S. Peck   ORCID: orcid.org/0000-0003-3479-6791 12 ,
• Ricardo Calado 13 ,
• Omar Defeo   ORCID: orcid.org/0000-0001-8318-528X 14 ,
• Steven Degraer   ORCID: orcid.org/0000-0002-3159-5751 15 ,
• Emma L. Johnston   ORCID: orcid.org/0000-0002-2117-366X 16 ,
• Hermanni Kaartokallio 17 ,
• Peter I. Macreadie   ORCID: orcid.org/0000-0001-7362-0882 18 ,
• Anna Metaxas   ORCID: orcid.org/0000-0002-1935-6213 19 ,
• Agnes W. N. Muthumbi 20 ,
• David O. Obura   ORCID: orcid.org/0000-0003-2256-6649 21 , 22 ,
• David M. Paterson 23 ,
• Alberto R. Piola   ORCID: orcid.org/0000-0002-5003-8926 24 , 25 ,
• Anthony J. Richardson   ORCID: orcid.org/0000-0002-9289-7366 26 , 27 ,
• Irene R. Schloss   ORCID: orcid.org/0000-0002-5917-8925 28 , 29 , 30 ,
• Paul V. R. Snelgrove   ORCID: orcid.org/0000-0002-6725-0472 31 ,
• Bryce D. Stewart 32 ,
• Paul M. Thompson   ORCID: orcid.org/0000-0001-6195-3284 33 ,
• Gordon J. Watson   ORCID: orcid.org/0000-0001-8274-7658 34 ,
• Thomas A. Worthington   ORCID: orcid.org/0000-0002-8138-9075 2 ,
• Moriaki Yasuhara   ORCID: orcid.org/0000-0003-0990-1764 35 &
• William J. Sutherland 2 , 36  

Nature Ecology & Evolution volume  6 ,  pages 1262–1270 ( 2022 ) Cite this article

25k Accesses

26 Citations

663 Altmetric

Metrics details

• Ocean sciences
• Science, technology and society
• Scientific community

The biodiversity of marine and coastal habitats is experiencing unprecedented change. While there are well-known drivers of these changes, such as overexploitation, climate change and pollution, there are also relatively unknown emerging issues that are poorly understood or recognized that have potentially positive or negative impacts on marine and coastal ecosystems. In this inaugural Marine and Coastal Horizon Scan, we brought together 30 scientists, policymakers and practitioners with transdisciplinary expertise in marine and coastal systems to identify new issues that are likely to have a significant impact on the functioning and conservation of marine and coastal biodiversity over the next 5–10 years. Based on a modified Delphi voting process, the final 15 issues presented were distilled from a list of 75 submitted by participants at the start of the process. These issues are grouped into three categories: ecosystem impacts, for example the impact of wildfires and the effect of poleward migration on equatorial biodiversity; resource exploitation, including an increase in the trade of fish swim bladders and increased exploitation of marine collagens; and new technologies, such as soft robotics and new biodegradable products. Our early identification of these issues and their potential impacts on marine and coastal biodiversity will support scientists, conservationists, resource managers and policymakers to address the challenges facing marine ecosystems.

Similar content being viewed by others

Well-being outcomes of marine protected areas

Natalie C. Ban, Georgina Grace Gurney, … Sara Jo Breslow

The Marine Spatial Planning Index: a tool to guide and assess marine spatial planning

Julie M. Reimer, Rodolphe Devillers, … Joachim Claudet

Marine ecosystem-based management: challenges remain, yet solutions exist, and progress is occurring

J. B. Haugen, J. S. Link, … A. L. Agnalt

The fifteenth Conference of the Parties (COP) to the United Nations Convention on Biological Diversity will conclude negotiations on a global biodiversity framework in late-2022 that will aim to slow and reverse the loss of biodiversity and establish goals for positive outcomes by 2050 1 . Currently recognized drivers of declines in marine and coastal ecosystems include overexploitation of resources (for example, fishes, oil and gas), expansion of anthropogenic activities leading to cumulative impacts on the marine and coastal environment (for example, habitat loss, introduction of contaminants and pollution) and effects of climate change (for example, ocean warming, freshening and acidification). Within these broad categories, marine and coastal ecosystems face a wide range of emerging issues that are poorly recognized or understood, each having the potential to impact biodiversity. Researchers, conservation practitioners and marine resource managers must identify, understand and raise awareness of these relatively ‘unknown’ issues to catalyse further research into their underlying processes and impacts. Moreover, informing the public and policymakers of these issues can mitigate potentially negative impacts through precautionary principles before those effects become realized: horizon scans provide a platform to do this.

Horizon scans bring together experts from diverse disciplines to discuss issues that are (1) likely to have a positive or negative impact on biodiversity and conservation within the coming years and (2) not well known to the public or wider scientific community or face a substantial ‘step-change’ in their importance or application 2 . Horizon scans are an effective approach for pre-emptively identifying issues facing global conservation 3 . Indeed, marine issues previously identified through this approach include microplastics 4 , invasive lionfish 4 and electric pulse trawling 5 . To date, however, no horizon scan of this type has focused solely on issues related to marine and coastal biodiversity, although a scan on coastal shorebirds in 2012 identified potential threats to coastal ecosystems 6 . This horizon scan aims to benefit our ocean and human society by stimulating research and policy development that will underpin appropriate scientific advice on prevention, mitigation, management and conservation approaches in marine and coastal ecosystems.

We present the final 15 issues below in thematic groups identified post-scoring, rather than rank order (Fig. 1 ).

Numbers refer to the order presented in this article, rather than final ranking. Image of brine pool courtesy of the NOAA Office of Ocean Exploration and Research, Gulf of Mexico 2014. Image of biodegradable bag courtesy of Katie Dunkley.

Ecosystem impacts

Wildfire impacts on coastal and marine ecosystems.

The frequency and severity of wildfires are increasing with climate change 7 . Since 2017, there have been fires of unprecedented scale and duration in Australia, Brazil, Portugal, Russia and along the Pacific coast of North America. In addition to threatening human life and releasing stored carbon, wildfires release aerosols, particles and large volumes of materials containing soluble forms of nutrients including nitrogen, phosphorus and trace metals such as copper, lead and iron. Winds and rains can transport these materials over long distances to reach coastal and marine ecosystems. Australian wildfires, for example, triggered widespread phytoplankton blooms in the Southern Ocean 8 along with fish and invertebrate kills in estuaries 9 . Predicting the magnitude and effects of these acute inputs is difficult because they vary with the size and duration of wildfires, the burning vegetation type, rainfall patterns, riparian vegetation buffers, dispersal by aerosols and currents, seasonal timing and nutrient limitation in the recipient ecosystem. Wildfires might therefore lead to beneficial, albeit temporary, increases in primary productivity, produce no effect or have deleterious consequences, such as the mortality of benthic invertebrates, including corals, from sedimentation, coastal darkening (see below), eutrophication or algal blooms 10 .

Coastal darkening

Coastal ecosystems depend on the penetration of light for primary production by planktonic and attached algae and seagrass. However, climate change and human activities increase light attenuation through changes in dissolved materials modifying water colour and suspended particles. Increased precipitation, storms, permafrost thawing and coastal erosion have led to the ‘browning’ of freshwater ecosystems by elevated organic carbon, iron and particles, all of which are eventually discharged into the ocean 11 . Coastal eutrophication leading to algal blooms compounds this darkening by further blocking light penetration. Additionally, land-use change, dredging and bottom fishing can increase seafloor disturbance, resuspending sediments and increasing turbidity. Such changes could affect ocean chemistry, including photochemical degradation of dissolved organic carbon and generation of toxic chemicals. At moderate intensities, limited spatial scales and during heatwaves, coastal darkening may have some positive impacts such as limiting coral bleaching on shallow reefs 12 but, at high intensities and prolonged spatial and temporal extents, lower light-regimes can contribute to cumulative stressor effects thereby profoundly altering ecosystems. This darkening may result in shifts in species composition, distribution, behaviour and phenology, as well as declines in coastal habitats and their functions (for example, carbon sequestration) 13 .

Increased toxicity of metal pollution due to ocean acidification

Concerns about metal toxicity in the marine environment are increasing as we learn more about the complex interactions between metals and global climate change 14 . Despite tight regulation of polluters and remediation efforts in some countries, the high persistence of metals in contaminated sediments results in the ongoing remobilization of existing metal pollutants by storms, trawling and coastal development, augmented by continuing release of additional contaminants into coastal waters, particularly in urban and industrial areas across the globe 14 . Ocean acidification increases the bioavailability, uptake and toxicity of metals in seawater and sediments, with direct toxicity effects on some marine organisms 15 . Not all biogeochemical changes will result in increased toxicity; in pelagic and deep-sea ecosystems, where trace metals are often deficient, increasing acidity may increase bioavailability and, in shallow waters, stimulate productivity for non-calcifying phytoplankton 16 . However, increased uptake of metals in wild-caught and farmed bivalves linked to ocean acidification could also affect human health, especially given that these species provide 25% of the world’s seafood. The combined effects of ocean acidification and metals could not only increase the levels of contamination in these organisms but could also impact their populations in the future 14 .

Equatorial marine communities are becoming depauperate due to climate migration

Climate change is causing ocean warming, resulting in a poleward shift of existing thermal zones. In response, species are tracking the changing ocean environmental conditions globally, with range shifts moving five times faster than on land 17 . In mid-latitudes and higher latitudes, as some species move away from current distribution ranges, other species from warmer regions can replace them 18 . However, the hottest climatic zones already host the most thermally tolerant species, which cannot be replaced due to their geographical position. Thus, climate change reduces equatorial species richness and has caused the formerly unimodal latitudinal diversity gradient in many communities to now become bimodal. This bimodality (dip in equatorial diversity) is projected to increase within the next 100 years if carbon dioxide emissions are not reduced 19 . The ecological consequences of this decline in equatorial zones are unclear, especially when combined with impacts of increasing human extraction and pollution 20 . Nevertheless, emerging ecological communities in equatorial systems are likely to have reduced resilience and capacity to support ecosystem services and human livelihoods.

Effects of altered nutritional content of fish due to climate change

Essential fatty acids (EFAs) are critical to maintaining human and animal health and fish consumption provides the primary source of EFAs for billions of people. In aquatic ecosystems, phytoplankton synthesize EFAs, such as docosahexaenoic acid (DHA) 21 , with pelagic fishes then consuming phytoplankton. However, concentrations of EFAs in fishes vary, with generally higher concentrations of omega-3 fatty acids in slower-growing species from colder waters 22 . Ongoing effects of climate change are impacting the production of EFAs by phytoplankton, with warming waters predicted to reduce the availability of DHA by about 10–58% by 2100 23 ; a 27.8% reduction in available DHA is associated with a 2.5 °C rise in water temperature 21 . Combined with geographical range shifts in response to environmental change affecting the abundance and distribution of fishes, this could lead to a reduction in sufficient quantities of EFAs for fishes, particularly in the tropics 24 . Changes to EFA production by phytoplankton in response to climate change, as shown for Antarctic waters 25 , could have cascading effects on the nutrient content of species further up the food web, with consequences for marine predators and human health 26 .

Resource exploitation

The untapped potential of marine collagens and their impacts on marine ecosystems.

Collagens are structural proteins increasingly used in cosmetics, pharmaceuticals, nutraceuticals and biomedical applications. Growing demand for collagen has fuelled recent efforts to find new sources that avoid religious constraints and alleviate risks associated with disease transmission from conventional bovine and porcine sources 27 . The search for alternative sources has revealed an untapped opportunity in marine organisms, such as from fisheries bycatch 28 . However, this new source may discourage efforts to reduce the capture of non-target species. Sponges and jellyfish offer a premium source of marine collagens. While the commercial-scale harvesting of sponges is unlikely to be widely sustainable, there may be some opportunity in sponge aquaculture and jellyfish harvesting, especially in areas where nuisance jellyfish species bloom regularly (for example, Mediterranean and Japan Seas). The use of sharks and other cartilaginous fish to supply marine collagens is of concern given the unprecedented pressure on these species. However, the use of coproducts derived from the fish-processing industry (for example, skin, bones and trims) offers a more sustainable approach to marine collagen production and could actively contribute to the blue bio-economy agenda and foster circularity 29 .

Impacts of expanding trade for fish swim bladders on target and non-target species

In addition to better-known luxury dried seafoods, such as shark fins, abalone and sea cucumbers, there is an increasing demand for fish swim bladders, also known as fish maw 30 . This demand may trigger an expansion of unsustainable harvests of target fish populations, with additional impacts on marine biodiversity through bycatch 30 , 31 . The fish swim-bladder trade has gained a high profile because the overexploitation of totoaba ( Totoaba macdonaldi) has driven both the target population and the vaquita ( Phocoena sinus) (which is bycaught in the Gulf of Mexico fishery) to near extinction 32 . By 2018, totoaba swim bladders were being sold for US$46,000 kg −1 . This extremely lucrative trade disrupts efforts to encourage sustainable fisheries. However, increased demand on the totoaba was itself caused by overexploitation over the last century of the closely related traditional species of choice, the Chinese bahaba ( Bahaba taipingensis) . We now risk both repeating this pattern and increasing its scale of impact, where depletion of a target species causes markets to switch to species across broader taxonomic and biogeographical ranges 31 . Not only does this cascading effect threaten other croakers and target species, such as catfish and pufferfish but maw nets set in more diverse marine habitats are likely to create bycatch of sharks, rays, turtles and other species of conservation concern.

Impacts of fishing for mesopelagic species on the biological ocean carbon pump

Growing concerns about food security have generated interest in harvesting largely unexploited mesopelagic fishes that live at depths of 200–1,000 m (ref. 33 ). Small lanternfishes (Myctophidae) dominate this potentially 10 billion ton community, exceeding the mass of all other marine fishes combined 34 and spanning millions of square kilometres of the open ocean. Mesopelagic fish are generally unsuitable for human consumption but could potentially provide fishmeal for aquaculture 34 or be used for fertilizers. Although we know little of their biology, their diel vertical migration transfers carbon, obtained by feeding in surface waters at night, to deeper waters during the day across many hundreds and even thousands of metres depth where it is released by excretion, egestion and death. This globally important carbon transport pathway contributes to the biological pump 35 and sequesters carbon to the deep sea 36 . Recent estimates put the contribution of all fishes to the biological ocean pump at 16.1% (± s.d. 13%) (ref. 37 ). The potential large-scale removal of mesopelagic fishes could disrupt a major pathway of carbon transport into the ocean depths.

Extraction of lithium from deep-sea brine pools

Global groups, such as the Deep-Ocean Stewardship Initiative, emphasize increasing concern about the ecosystem impacts from deep-sea resource extraction 38 . The demand for batteries, including for electric vehicles, will probably lead to a demand for lithium that is more than five times its current level by 2030 39 . While concentrations are relatively low in seawater, some deep-sea brines and cold seeps offer higher concentrations of lithium. Furthermore, new technologies, such as solid-state electrolyte membranes, can enrich the concentration of lithium from seawater sources by 43,000 times, increasing the energy efficiency and profitability of lithium extraction from the sea 39 . These factors could divert extraction of lithium resources away from terrestrial to marine mining, with the potential for significant impacts to localized deep-sea brine ecosystems. These brine pools probably host many endemic and genetically distinct species that are largely undiscovered or awaiting formal description. Moreover, the extremophilic species in these environments offer potential sources of marine genetic resources that could be used in new biomedical applications including pharmaceuticals, industrial agents and biomaterials 40 . These concerns point to the need to better quantify and monitor biodiversity in these extreme environments to establish baselines and aid management.

New technologies

Colocation of marine activities.

Climate change, energy needs and food security have moved to the top of global policy agendas 41 . Increasing energy needs, alongside the demands of fisheries and transport infrastructure, have led to the proposal of colocated and multifunctional structures to deliver economic benefits, optimize spatial planning and minimize the environmental impacts of marine activities 42 . These designs often bring technical, social, economic and environmental challenges. Some studies have begun to explore these multipurpose projects (for example, offshore windfarms colocated with aquaculture developments and/or Marine Protected Areas) and how to adapt these concepts to ensure they are ‘fit for purpose’, economically viable and reliable. However, environmental and ecosystem assessment, management and regulatory frameworks for colocated and multi-use structures need to be established to prevent these activities from compounding rather than mitigating the environmental impacts from climate change 43 .

Floating marine cities

In April 2019, the UN-HABITAT programme convened a meeting of scientists, architects, designers and entrepreneurs to discuss how floating cities might be a solution to urban challenges such as climate change and lack of housing associated with a rising human population ( https://unhabitat.org/roundtable-on-floating-cities-at-unhq-calls-for-innovation-to-benefit-all ). The concept of floating marine cities—hubs of floating structures placed at sea—was born in the middle of the twentieth century and updated designs now aim to translate this vision into reality 44 . Oceanic locations provide benefits from wave and tidal renewable energy and food production supported by hydroponic agriculture 45 . Modular designs also offer greater flexibility than traditional static terrestrial cities, whereby accommodation and facilities could be incorporated or removed in response to changes in population or specific events. The cost of construction in harsh offshore environments, rather than technology, currently limits the development of marine cities and potential designs will need to consider the consequences of more frequent and extreme climate events. Although the artificial hard substrates created for these floating cities could act as stepping stones, facilitating species movement in response to climate change 46 , this could also increase the spread of invasive species. Finally, the development of offshore living will raise issues in relation to governance and land ownership that must be addressed for marine cities to be viable 47 .

Trace-element contamination compounded by the global transition to green technologies

The persistent environmental impacts of metal and metalloid trace-element contamination in coastal sediments are now increasing after a long decline 48 . However, the complex sources of contamination challenge their management. The acceleration of the global transition to green technologies, including electric vehicles, will increase demand for batteries by over 10% annually in the coming years 49 . Electric vehicle batteries currently depend almost exclusively on lithium-ion chemistries, with potential trace-element emissions across their life cycle from raw material extraction to recycling or end-of-life disposal. Few jurisdictions treat lithium-ion batteries as harmful waste, enabling landfill disposal with minimal recycling 49 . Cobalt and nickel are the primary ecotoxic elements in next-generation lithium-ion batteries 50 , although there is a drive to develop a cobalt-free alternative likely to contain higher nickel content 50 . Some battery binder and electrolyte chemicals are toxic to aquatic life or form persistent organic pollutants during incomplete burning. Increasing pollution from battery production, recycling and disposal in the next decade could substantially increase the potentially toxic trace-element contamination in marine and coastal systems worldwide.

New underwater tracking systems to study non-surfacing marine animals

The use of tracking data in science and conservation has grown exponentially in recent decades. Most trajectory data collected on marine species to date, however, has been restricted to large and near-surface species, limited by the size of the devices and reliance on radio signals that do not propagate well underwater. New battery-free technology based on acoustic telemetry, named ‘underwater backscatter localization’ (UBL), may allow high-accuracy (<1 m) tracking of animals travelling at any depth and over large distances 51 . Still in the early stages of development, UBL technology has significant potential to help fill knowledge gaps in the distribution and spatial ecology of small, non-surfacing marine species, as well as the early life-history stages of many species 52 , over the next decades. However, the potential negative impacts of this methodology on the behaviour of animals are still to be determined. Ultimately, UBL may inform spatial management both in coastal and offshore regions, as well as in the high seas and address a currently biased perspective of how marine animals use ocean space, which is largely based on near-surface or aerial marine megafauna (for example, ref. 53 ).

Soft robotics for marine research

The application and utility of soft robotics in marine environments is expected to accelerate in the next decade. Soft robotics, using compliant materials inspired by living organisms, could eventually offer increased flexibility at depth because they do not face the same constraints as rigid robots that need pressurized systems to function 54 . This technology could increase our ability to monitor and map the deep sea, with both positive and negative consequences for deep-sea fauna. Soft-grab robots could facilitate collection of delicate samples for biodiversity monitoring but, without careful management, could also add pollutants and waste to these previously unexplored and poorly understood environments 55 . With advancing technology, potential deployment of swarms of small robots could collect basic environmental data to facilitate mapping of the seabed. Currently limited by power supply, energy-harvesting modules are in development that enable soft robots to ‘swallow’ organic material and convert it into power 56 , although this could result in inadvertently harvesting rare deep-sea organisms. Soft robots themselves may also be ingested by predatory species mistaking them for prey. Deployment of soft robotics will require careful monitoring of both its benefits and risks to marine biodiversity.

The effects of new biodegradable materials in the marine environment

Mounting public pressure to address marine plastic pollution has prompted the replacement of some fossil fuel-based plastics with bio-based biodegradable polymers. This consumer pressure is creating an economic incentive to adopt such products rapidly and some companies are promoting their environmental benefits without rigorous toxicity testing and/or life-cycle assessments. Materials such as polybutylene succinate (PBS), polylactic acid (PLA) or cellulose and starch-based materials may become marine litter and cause harmful effects akin to conventional plastics 57 . The long-term and large-scale effect of the use of biodegradable polymers in products (for example, clothing) and the unintended release of byproducts, such as microfibres, into the environment remain unknown. However, some natural microfibres have greater toxicity than plastic microfibres when consumed by aquatic invertebrates 58 . Jurisdictions should enact and enforce suitable regulations to require the individual assessment of all new materials intended to biodegrade in a full range of marine environmental conditions. In addition, testing should include studies on the toxicity of major transition chemicals created during the breakdown process 59 , ideally considering the different trophic levels of marine food webs.

This scan identified three categories of horizon issues: impacts on, and alterations to, ecosystems; changes to resource use and extraction; and the emergence of technologies. While some of the issues discussed, such as improved monitoring of species (underwater tracking and soft robotics) and more sustainable resource use (marine collagens), may have some positive outcomes for marine and coastal biodiversity, most identified issues are expected to have substantial negative impacts if not managed or mitigated appropriately. This imbalance highlights the considerable emerging pressures facing marine ecosystems that are often a byproduct of human activities.

Four issues identified in this scan related to ongoing large-scale (hundreds to many thousands of square kilometres) alterations to marine ecosystems (wildfires, coastal darkening, depauperate equatorial communities and altered nutritional fish content), either through the impacts of global climate change or other human activities. There are already clear impacts of climate change, for example, on stores of blue carbon (for example, ref. 60 ) and small-scale fisheries (for example, ref. 61 ) but the identification of these issues highlights the need for global action that reverses such trends. The United Nations Decade of Ocean Science for Sustainable Development (2021–2030) is now underway, aligning with other decadal policy priorities, including the Sustainable Development Goals ( https://sdgs.un.org/ ), the 2030 targets for biodiversity to be agreed in 2022, the conclusion of the ongoing negotiations on biodiversity beyond national jurisdictions (BBNJ) ( https://www.un.org/bbnj/ ), the UN Conference on Biodiversity (COP15) ( https://www.unep.org/events/conference/un-biodiversity-conference-cop-15 ) and the UN Climate Change Conference 2021 (COP26) ( https://ukcop26.org/ ). While some campaigns to allocate 30% of the ocean to Marine Protected Areas by 2030 are prominently aired 62 , the unintended future consequences of such protection and how to monitor and manage these areas, remain unclear 63 , 64 , 65 .

Another set of issues related to anticipated increases in marine resource use and extraction (swim bladders, marine collagens, lithium extraction and mesopelagic fisheries). The complex issue of mitigating the impacts on marine conservation and biodiversity of exploiting and using newly discovered resources must consider public perceptions of the ocean 66 , 67 , market forces and the sustainable blue economy 68 , 69 .

The final set of issues related to new technological advancements, with many offering more sustainable opportunities, albeit some having potentially unintended negative consequences on marine and coastal biodiversity. For example, trace-element contamination from green technologies and harmful effects of biodegradable products highlights the need to assess the step-changes in impacts from their increased use and avoid the paradox of technologies designed to mitigate the damaging effects of climate change on biodiversity themselves damaging biodiversity. Indeed, the impacts on marine and coastal biodiversity from emerging technologies currently in development (such as underwater tracking or soft robotics) need to be assessed before deployment at scale.

There are limitations to any horizon scanning process that aims to identify global issues and a different group of experts may have identified a different set of issues. By inviting participants from a range of subject backgrounds and global regions and asking them to canvass their network of colleagues and collaborators, we aimed to identify as broad a set of issues as possible. We acknowledge, however, that only about one-quarter of the participants were from non-academic organizations, which may have skewed the submitted issues and how they were voted on. However, others 3 reported no significant correlation between participants’ areas of research expertise and the top issues selected in the horizon scan conducted in 2009. Therefore, horizon scans do not necessarily simply represent issues that reflect the expertise of participants. We also sought to achieve diversity by inviting participants from 22 countries and actively seeking representatives from the global south. However, the final panel of 30 participants spanned only 11 countries, most in the global north. We were forced by the COVID-19 pandemic to hold the scan online and while we hoped that this would enable participants to engage from around the world alleviating broader global inequalities in science 63 , digital inequality was in fact enhanced during the pandemic 70 . Our experience highlights the need for other mechanisms that can promote global representation in these scans.

This Marine and Coastal Horizon Scan seeks to raise awareness of issues that may impact marine and coastal biodiversity conservation in the next 5–10 years. Our aim is to bring these issues to the attention of scientists, policymakers, practitioners and the wider community, either directly, through social networks or the mainstream media. Whilst it is almost impossible to determine whether issues gained prominence as a direct result of a horizon scan, some issues featured in previous scans have seen growth in reporting and awareness. Others 3 found that 71% of topics identified in the Horizon Scan in 2009 had seen an increase in their importance over the next 10 years. Issues such as microplastics and invasive lionfish had received increased research and investment from scientists, funders, managers and policymakers to understand their impacts and the horizon scans may have helped motivate this increase. Horizon scans, therefore, should primarily act as signposts, putting focus onto particular issues and providing support for researchers and practitioners to seek investment in these areas.

Whilst recognizing that marine and coastal environments are complex social-ecological systems, the role of governance, policy and litigation on all areas of marine science needs to be developed, as it is yet to be established to the same extent as in terrestrial ecosystems 71 . Indeed, tackling many of the issues presented in this scan will require an understanding of the human dimensions relating to these issues, through fields of research including but not limited to ocean literacy 72 , 73 , social justice, equity 74 and human health 75 . Importantly, however, horizon scanning has proved an efficient tool in identifying issues that have subsequently come to the forefront of public knowledge and policy decisions, while also helping to focus future research. The scale of the issues facing marine and coastal areas emphasizes the need to identify and prioritize, at an early stage, those issues specifically facing marine ecosystems, especially within this UN Decade of Ocean Science for Sustainable Development.

Identification of issues

In March 2021, we brought together a core team of 11 participants from a broad range of marine and coastal disciplines. The core team suggested names of individuals outside their subject area who were also invited to participate in the horizon scan. To ensure we included as many different subject areas as possible within marine and coastal conservation, we selected one individual from each discipline. Our panel of experts comprised 30 (37% female) marine and coastal scientists, policymakers and practitioners (27% from non-academic institutions), with cross-disciplinary expertise in ecology (including tropical, temperate, polar and deep-sea ecosystems), palaeoecology, conservation, oceanography, climate change, ecotoxicology, technology, engineering and marine social sciences (including governance, blue economy and ocean literacy). Participants were invited from 22 countries across six continents, resulting in a final panel of 30 experts from 11 countries (Europe n  = 17 (including the three organizers); North America and Caribbean n  = 4; South America n  = 3; Australasia n  = 3; Asia n  = 1; Africa n  = 2). All experts co-authored this paper.

To reduce the potential for bias in the identification of suitable issues, each participant was invited to consult their own network and required to submit two to five issues that they considered new and likely to have a positive or negative impact on marine and coastal biodiversity conservation in the next 5–10 years ( Supplementary Information text describes instructions given to participants). Each issue was described in paragraphs of ~200 words (plus references). Due to the COVID-19 pandemic, participants relied mainly on virtual meetings and online communication using email, social-media platforms, online conferences and networking events. Through these channels ~680 people were canvassed by the participants, counting all direct in-person or online discussions as individual contacts but treating social-media posts or generic emails as a single contact. This process resulted in a long list of 75 issues that were considered in the first round of scoring (see Supplementary Table 1 for the full list of initially submitted issues).

Round 1 scoring

The initial list of proposed issues was then shortened through a scoring process. We used a modified Delphi-style 76 voting process, which has been consistently applied in horizon scans since 2009 (refs. 4 , 77 ) (see Fig. 2 for the stepwise process). This process ensured that consideration and selection of issues remained repeatable, transparent and inclusive. Panel members were asked to confidentially and independently score the long list of 75 issues from 1 (low) to 1,000 (high) on the basis of the following criteria:

Whether the issue is new (with ‘new’ issues scoring higher) or is a well-known issue likely to exhibit a significant step-change in impact

Whether the issue is likely to be important and impactful over the next 5–10 years

Whether the issue specifically impacts marine and coastal biodiversity

Left and right columns show the process for the first and second rounds of scoring, respectively.

Participants were also asked whether they had heard of the issue or not.

‘Voter fatigue’ can result in issues at the end of a lengthy list not receiving the same consideration as those at the beginning 76 . We counteracted this potential bias by randomly assigning participants to one of three differently ordered long-lists of issues. Participants’ scores were converted to ranks (1–75). We had aimed to retain the top 30 issues with the highest median ranks for the second round of assessment at the workshop but kept 31 issues because two issues achieved equal median ranks. In addition, we identified one issue that had been incorrectly grouped with three others and presented this as a separate issue. The subsequent online workshop to discuss this shortlist, therefore, considered the top-ranked 32 issues (Fig. 3a ) (see Supplementary Table 2 for the full list).

a , Round 1. Each point represents an individual issue. For all issue titles, see Supplementary Table 1 . Issues in dark blue were retained for the second round. Issues that were ranked higher were generally those that participants had not heard of (Spearman rank correlation = 0.38, P  < 0.001). b , Round 2. Scores as in round 1. For titles of the second round of 32 issues, see Supplementary Table 2 . The 15 final issues (marked in red) achieved the top ranks (horizontal dashed line) and had only been heard of by 50% of participants (vertical dashed line). Red circles, squares and triangles denote issues relating to ecosystem impacts, resource exploitation and new technologies, respectively. The two grey issues marked with crosses were discounted during final discussions because participants could not identify the horizon component of these issues.

Source data

Workshop and round 2 scoring.

Before the workshop, each participant was assigned up to four of the 32 issues to research in more detail and contribute further information to the discussion. We convened a one-day workshop online in September 2021. The geographic spread of participants meant that time zones spanned 17 h. Despite these constraints, discussions remained detailed, focused, varied and lively. In addition, participants made use of the chat function on the platform to add notes, links to articles and comments to the discussion. After discussing each issue, participants re-scored the topic (1–1,000, low to high) based on novelty and the issue’s importance for, and probable impact on, marine and coastal biodiversity (3 participants out of 30 did not score all issues and therefore their scores were discounted). At the end of the selection process, scores were again converted to ranks and collated. Highest-ranked issues were then discussed by correspondence focusing on the same three criteria as outlined above, after which the top 15 horizon issues were selected (Fig. 3b ).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability

The datasets generated during and/or analysed during the current study are available from figshare https://doi.org/10.6084/m9.figshare.19703485.v1 . Source data are provided with this paper.

Díaz, S. et al. Set ambitious goals for biodiversity and sustainability. Science 370 , 411–413 (2020).

Article   PubMed   Google Scholar  

Sutherland, W. J. & Woodroof, H. J. The need for environmental horizon scanning. Trends Ecol. Evol. 24 , 523–527 (2009).

Sutherland, W. J. et al. Ten years on: a review of the first global conservation horizon scan. Trends Ecol. Evol. 34 , 139–153 (2019).

Sutherland, W. J. et al. A horizon scan of global conservation issues for 2010. Trends Ecol. Evol. 25 , 1–7 (2010).

Sutherland, W. J. et al. A horizon scan of global conservation issues for 2016. Trends Ecol. Evol. 31 , 44–53 (2016).

Sutherland, W. J. et al. A horizon scanning assessment of current and potential future threats facing migratory shorebirds. Ibis 154 , 663–679 (2012).

Article   Google Scholar  

Bowman, D. M. J. S. et al. Vegetation fires in the Anthropocene. Nat. Rev. Earth Environ. 1 , 500–515 (2020).

Tang, W. et al. Widespread phytoplankton blooms triggered by 2019–2020 Australian wildfires. Nature 597 , 370–375 (2021).

Article   CAS   PubMed   Google Scholar  

Silva, L. G. M. et al. Mortality events resulting from Australia’s catastrophic fires threaten aquatic biota. Glob. Change Biol. 26 , 5345–5350 (2020).

Abram, N. J., Gagan, M. K., McCulloch, M. T., Chappell, J. & Hantoro, W. S. Coral reef death during the 1997 Indian Ocean Dipole linked to Indonesian wildfires. Science 301 , 952–955 (2003).

Solomon, C. T. et al. Ecosystem consequences of changing inputs of terrestrial dissolved organic matter to lakes: current knowledge and future challenges. Ecosystems 18 , 376–389 (2015).

Sully, S. & van Woesik, R. Turbid reefs moderate coral bleaching under climate related temperature stress. Glob. Change Biol. 26 , 1367–1373 (2021).

Blain, C. O., Hansen, S. C. & Shears, N. T. Coastal darkening substantially limits the contribution of kelp to coastal carbon cycles. Glob. Change Biol. 27 , 5547–5563 (2021).

Article   CAS   Google Scholar  

Stewart, B. D. et al. Metal pollution as a potential threat to shell strength and survival in marine bivalves. Sci. Total Environ. 755 , 143019 (2021).

Roberts, D. A. et al. Ocean acidification increases the toxicity of contaminated sediments. Glob. Change Biol. 19 , 340–351 (2013).

Hauton, C. et al. Identifying toxic impact of metals potentially released during deep-sea mining—a synthesis of the challenges to quantifying risk. Front. Mar. Sci . 4 , 368 (2017).

Chaudhary, C. et al. Global warming is causing a more pronounced dip in marine species richness around the equator. Proc. Natl Acad. Sci. USA 118 , e2015094118 (2021).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Burrows, M. T. et al. Geographical limits to species-range shifts are suggested by climate velocity. Nature 507 , 492–495 (2014).

Yasuhara, M. et al. Past and future decline of tropical pelagic biodiversity. Proc. Natl Acad. Sci. USA 117 , 12891–12896 (2020).

Pandolfi, J. M. et al. Are U.S. coral reefs on the slippery slope to slime? Science 307 , 1725–1726 (2005).

Hixson, S. M. & Arts, M. T. Climate warming is predicted to reduce omega-3, long-chain, polyunsaturated fatty acid production in phytoplankton. Glob. Change Biol. 22 , 2744–2755 (2016).

Hicks, C. C. et al. Harnessing global fisheries to tackle micronutrient deficiencies. Nature 574 , 95–98 (2019).

Colombo, S. M. et al. Projected declines in global DHA availability for human consumption as a result of global warming. Ambio 49 , 865–880 (2020).

Lam, V. W. et al. Climate change, tropical fisheries and prospects for sustainable development. Nat. Rev. Earth Environ. 1 , 440–454 (2020).

Antacli, J. C. et al. Increase in unsaturated fatty acids in Antarctic phytoplankton under ocean warming and glacial melting scenarios. Sci. Total Environ. 790 , 147879 (2021).

Maire, E. et al. Micronutrient supply from global marine fisheries under climate change and overfishing. Curr. Biol. 18 , 4132–4138 (2021).

Lim, Y. S., Ok, Y. J., Hwang, S. Y., Kwak, J. Y. & Yoon, S. Marine collagen as a promising biomaterial for biomedical applications. Mar. Drugs 17 , 467 (2019).

Article   CAS   PubMed Central   Google Scholar  

Xu, N. et al. Marine-derived collagen as biomaterials for human health. Front. Nutr. 8 , 702108 (2021).

Article   PubMed   PubMed Central   CAS   Google Scholar  

Vieira, H., Leal, M. C. & Calado, R. Fifty shades of blue: how blue biotechnology is shaping the bioeconomy. Trends Biotechnol. 38 , 940–943 (2020).

Ben-Hasan, A. et al. China’s fish maw demand and its implications for fisheries in source countries. Mar. Policy 132 , 104696 (2021).

Sadovy de Mitcheson, Y., To, A. W. L., Wong, N. W., Kwan, H. Y. & Bud, W. S. Emerging from the murk: threats, challenges and opportunities for the global swim bladder trade. Rev. Fish. Biol. Fish. 29 , 809–835 (2019).

Brownell, R. L. Jr et al. Bycatch in gillnet fisheries threatens critically endangered small cetaceans and other aquatic megafauna. Endang. Species Res. 40 , 285–296 (2019).

Webb, T. J., Vanden Berghe, E. & O’Dor, R. K. Biodiversity’s big wet secret: the global distribution of marine biological records reveals chronic under-exploration of the deep pelagic ocean. PLoS ONE 5 , e10223 (2010).

St. John, M. A. et al. A dark hole in our understanding of marine ecosystems and their services: perspectives from the mesopelagic community. Front. Mar. Sci. 3 , 31 (2016).

Google Scholar  

Thomsen, L. et al. The oceanic biological pump: rapid carbon transfer to depth at continental margins during winter. Sci. Rep. 7 , 10763 (2017).

Roberts, C. M., Hawkins, J. P., Hindle, K., Wilson, R. W. & O’Leary, B. C. Entering the Twilight Zone: The Ecological Role and Importance of Mesopelagic Fishes (Blue Marine Foundation, 2020)

Cavan, E. L., Laurenceau-Cornec, E. C., Bressac, M. & Boyd, P. W. Exploring the ecology of the mesopelagic biological pump. Prog. Oceanogr. 176 , 102125 (2019).

Levin, L. A. et al. Climate change considerations are fundamental to management of deep‐sea resource extraction. Glob. Change Biol. 26 , 4664–4678 (2020).

Li, Z. et al. Continuous electrical pumping membrane process for seawater lithium mining. Energy Environ. Sci. 14 , 3152–3159 (2021).

Jin, M., Gai, Y., Guo, X., Hou, Y. & Zeng, R. Properties and applications of extremozymes from deep-sea extremophilic microorganisms: a mini review. Mar. Drugs 17 , 656 (2019).

Mbow, C. et al. in IPCC Special Report on Climate Change and Land (eds Shukla, P.R. et al.) 437–550 (IPCC, 2019).

Christie, N., Smyth, K., Barnes, R. & Elliott, M. Co-location of activities and designations: a means of solving or creating problems in marine spatial planning? Mar. Pol. 43 , 254–261 (2014).

Mayer-Pinto, M., Dafforn, K. A. & Johnston, E. L. A decision framework for coastal infrastructure to optimize biotic resistance and resilience in a changing climate. BioScience 69 , 833–843 (2019).

Wang, C. M. & Wang, B. T. in ICSCEA 2019 (eds Reddy, J. N. et al.) 3–29 (Springer, 2020).

Ross, C. T. F. & McCullough, R. R. Conceptual design of a floating island city. J. Ocean Technol. 5 , 120–121 (2010).

Dong, Y.-w, Huang, X.-w, Wang, W., Li, Y. & Wang, J. The marine ‘great wall’ of China: local- and broad-scale ecological impacts of coastal infrastructure on intertidal macrobenthic communities. Divers. Distrib. 22 , 731–744 (2016).

Flikkema, M. M. B., Lin, F.-Y., van der Plank, P. P. J., Koning, J. & Waals, O. Legal issues for artificial floating islands. Front. Mar. Sci. 8 , 619462 (2021).

Richir, J., Bray, S., McAleese, T. & Watson, G. J. Three decades of trace element sediment contamination: the mining of governmental databases and the need to address hidden sources for clean and healthy seas. Environ. Int. 149 , 106362 (2021).

Zhao, Y. et al. A review on battery market trends, second-life reuse, and recycling. Sustain. Chem. 2 , 167–205 (2021).

Li, W., Lee, S. & Manthiram, A. High‐Nickel NMA: a cobalt‐free alternative to NMC and NCA cathodes for lithium‐ion batteries. Adv. Mater. 32 , 2002718 (2020).

Ghaffarivardavagh, R., Afzal, S. S., Rodriguez, O. & Adib, F. in SIGCOMM ’20 Proc. 19th ACM Workshop on Hot Topics in Networks 125–131 (Association for Computing Machinery, 2020).

Hazen, E. L. et al. Ontogeny in marine tagging and tracking science: technologies and data gaps. Mar. Ecol. Prog. Ser. 457 , 221–240 (2012).

Davies, T. E. et al. Tracking data and the conservation of the high seas: opportunities and challenges. J. Appl. Ecol . 58 , 2703–2710 (2021).

Aracri, S. et al. Soft robots for ocean exploration and offshore operations: a perspective. Soft Robot. https://doi.org/10.1089/soro.2020.0011 (2021).

Li, G. et al. Self-powered soft robot in the Mariana Trench. Nature 591 , 66–71 (2021).

Philamore, H., Ieropoulos, I., Stinchcombe, A. & Rossiter, J. Toward energetically autonomous foraging soft robots. Soft Robot. 3 , 186–197 (2016).

Manfra, L. et al. Biodegradable polymers: a real opportunity to solve marine plastic pollution? J. Hazard. Mater. 416 , 125763 (2021).

Kim, D., Kim, H. & An, Y. J. Effects of synthetic and natural microfibers on Daphnia magna : are they dependent on microfiber type? Aquat. Toxicol. 240 , 105968 (2021).

Degli-Innocenti, F., Bellia, G., Tosin, M., Kapanen, A. & Itävaara, M. Detection of toxicity released by biodegradable plastics after composting in activated vermiculite. Polym. Degrad. Stab. 73 , 101–106 (2001).

Macreadie, P. I. et al. The future of blue carbon science. Nat. Commun. 10 , 3998 (2019).

Short, R. E. et al. Harnessing the diversity of small-scale actors is key to the future of aquatic food systems. Nat. Food 2 , 733–741 (2021).

Watson, J. E. M. et al. Set a global target for ecosystems. Nature 578 , 360–362 (2020).

Obura, D. O. et al. Integrate biodiversity targets from local to global levels. Science 373 , 746 (2021).

Barnes, M. D., Glew, L., Wyborn, C. & Craigie, I. D. Prevent perverse outcomes from global protected area policy. Nat. Ecol. Evol. 2 , 759–762 (2018).

Grorud-Colvert, K. et al. The MPA Guide: a framework to achieve global goals for the ocean. Science 373 , eabf0861 (2021).

Jefferson, R. L., McKinley, E., Griffin, H., Nimmo, A. & Fletcher, S. Public perceptions of the ocean: lessons for marine conservation from a global research review. Front. Mar. Sci . 8 , 711245 (2021).

Potts, T., Pita, C., O’Higgins, T. & Mee, L. Who cares? European attitudes towards marine and coastal environments. Mar. Pol. 72 , 59–66 (2016).

Bennett, N. J. et al. Towards a sustainable and equitable blue economy. Nat. Sustain. 2 , 991–993 (2019).

Jouffray, J.-B., Blasiak, R., Norström, A. V., Österblom, H. & Nyström, M. The blue acceleration: the trajectory of human expansion into the ocean. One Earth 2 , 43–54 (2020).

Zheng, Y. & Walsham, G. Inequality of what? An intersectional approach to digital inequality under Covid-19. Inf. Organ. 31 , 100341 (2021).

Blythe, J. L., Armitage, D., Bennett, N. J., Silver, J. J. & Song, A. M. The politics of ocean governance transformations. Front. Mar. Sci. 8 , 634718 (2021).

Brennan, C., Ashley, M. & Molloy, O. A system dynamics approach to increasing ocean literacy. Front. Mar. Sci. 6 , 360 (2019).

Stoll-Kleemann, S. Feasible options for behavior change toward more effective ocean literacy: a systematic review. Front. Mar. Sci. 6 , 273 (2019).

Bennett, N. J. et al. Advancing social equity in and through marine conservation. Front. Mar. Sci. 8 , 711538 (2021).

Short, R. E. et al. Review of the evidence for oceans and human health relationships in Europe: a systematic map. Environ. Int. 146 , 106275 (2021).

Mukherjee, N. et al. The Delphi technique in ecology and biological conservation: applications and guidelines. Methods Ecol. Evol. 6 , 1097–1109 (2015).

Sutherland, W. J. et al. A 2021 horizon scan of emerging global biological conservation issues. Trends Ecol. Evol. 36 , 87–97 (2021).

Download references

Acknowledgements

This Marine and Coastal Horizon Scan was funded by Oceankind. S.N.R.B. is supported by EcoStar (DM048) and Cefas (My time). R.C. acknowledges FCT/MCTES for the financial support to CESAM (UIDP/50017/2020, UIDB/50017/2020, LA/P/0094/2020) through national funds. O.D. is supported by CSIC Uruguay and Inter-American Institute for Global Change Research. J.E.H.-R. is supported by the Whitten Lectureship in Marine Biology. S.A.K. is supported by a Natural Environment Research Council grant (NE/S00050X/1). P.I.M. is supported by an Australian Research Council Discovery Grant (DP200100575). D.M.P. is supported by the Marine Alliance for Science and Technology for Scotland (MASTS). A.R.P. is supported by the Inter-American Institute for Global Change Research. W.J.S. is funded by Arcadia. A.T. is supported by Oceankind. M.Y. is supported by the Deep Ocean Stewardship Initiative and bioDISCOVERY. We are grateful to everyone who submitted ideas to the exercise and the following who are not authors but who suggested a topic that made the final list: R. Brown (colocation of marine activities), N. Graham and C. Hicks (altered nutritional content of fish), A. Thornton (soft robotics), A. Vincent (fish swim bladders) and T. Webb (mesopelagic fisheries).

Author information

These authors contributed equally: James E. Herbert-Read, Ann Thornton.

Authors and Affiliations

Department of Zoology, University of Cambridge, Cambridge, UK

James E. Herbert-Read

Conservation Science Group, Department of Zoology, Cambridge University, Cambridge, UK

Ann Thornton, Thomas A. Worthington & William J. Sutherland

SpeSeas, D’Abadie, Trinidad and Tobago

Diva J. Amon

Marine Science Institute, University of California, Santa Barbara, CA, USA

The Centre for Environment, Fisheries and Aquaculture Science (Cefas), Lowestoft, UK

Silvana N. R. Birchenough

Department of Biological Sciences, Simon Fraser University, Burnaby, British Columbia, Canada

Isabelle M. Côté

Centre for Ecology, Evolution and Environmental Changes (cE3c), Department of Animal Biology, Faculdade de Ciências da Universidade de Lisboa, Lisbon, Portugal

Maria P. Dias

BirdLife International, The David Attenborough Building, Cambridge, UK

Centre for Ecology and Conservation, University of Exeter, Penryn, UK

Brendan J. Godley

Lancaster Environment Centre, Lancaster University, Lancaster, UK

Sally A. Keith

School of Earth and Environmental Sciences, Cardiff University, Cardiff, UK

Emma McKinley

British Antarctic Survey, Natural Environment Research Council, Cambridge, UK

Lloyd S. Peck

ECOMARE, CESAM—Centre for Environmental and Marine Studies, Department of Biology, University of Aveiro, Santiago University Campus, Aveiro, Portugal

Ricardo Calado

Laboratory of Marine Sciences (UNDECIMAR), Faculty of Sciences, University of the Republic, Montevideo, Uruguay

Royal Belgian Institute of Natural Sciences, Operational Directorate Natural Environment, Marine Ecology and Management, Brussels, Belgium

Steven Degraer

School of Biological, Earth, and Environmental Sciences, University of New South Wales, Sydney, New South Wales, Australia

Emma L. Johnston

Finnish Environment Institute, Helsinki, Finland

Hermanni Kaartokallio

Centre for Integrative Ecology, School of Life and Environmental Sciences, Deakin University, Burwood Campus, Burwood, Victoria, Australia

Peter I. Macreadie

Department of Oceanography, Dalhousie University, Halifax, Nova Scotia, Canada

Anna Metaxas

Department of Biology, University of Nairobi, Nairobi, Kenya

Agnes W. N. Muthumbi

Coastal Oceans Research and Development in the Indian Ocean, Mombasa, Kenya

David O. Obura

School of Biological Sciences, University of Queensland, St Lucia, Brisbane, Queensland, Australia

Scottish Oceans Institute, School of Biology, University of St Andrews, St Andrews, UK

David M. Paterson

Servício de Hidrografía Naval, Buenos Aires, Argentina

Alberto R. Piola

Instituto Franco-Argentino sobre Estudios de Clima y sus Impactos, CONICET/CNRS, Universidad de Buenos Aires, Buenos Aires, Argentina

School of Mathematics and Physics, The University of Queensland, St Lucia, Brisbane, Queensland, Australia

Anthony J. Richardson

Commonwealth Scientific and Industrial Research Organisation (CSIRO) Oceans and Atmosphere, Queensland Biosciences Precinct, St Lucia, Brisbane, Queensland, Australia

Instituto Antártico Argentino, Buenos Aires, Argentina

Irene R. Schloss

Centro Austral de Investigaciones Científicas (CADIC-CONICET), Ushuaia, Argentina

Universidad Nacional de Tierra del Fuego, Antártida e Islas del Atlántico Sur, Ushuaia, Argentina

Department of Ocean Sciences and Biology Department, Memorial University, St John’s, Newfoundland and Labrador, Canada

Paul V. R. Snelgrove

Department of Environment and Geography, University of York, York, UK

Bryce D. Stewart

Lighthouse Field Station, School of Biological Sciences, University of Aberdeen, Cromarty, UK

Paul M. Thompson

Institute of Marine Sciences, School of Biological Sciences, University of Portsmouth, Portsmouth, UK

Gordon J. Watson

School of Biological Sciences, Area of Ecology and Biodiversity, Swire Institute of Marine Science, Institute for Climate and Carbon Neutrality, Musketeers Foundation Institute of Data Science, and State Key Laboratory of Marine Pollution, The University of Hong Kong, Kadoorie Biological Sciences Building, Hong Kong, China

Moriaki Yasuhara

Biosecurity Research Initiative at St Catharine’s (BioRISC), St Catharine’s College, University of Cambridge, Cambridge, UK

William J. Sutherland

You can also search for this author in PubMed   Google Scholar

Contributions

J.E.H.-R. and A.T. contributed equally to the manuscript. J.E.H.-R., A.T. and W.J.S. devised, organized and led the Marine and Coastal Horizon Scan. D.J.A., S.N.R.B., I.M.C., M.P.D., B.J.G., S.A.K., E.M. and L.S.P. formed the core team and are listed alphabetically in the author list. All other authors, R.C., O.D., S.D., E.L.J., H.K., P.I.M., A.M., A.W.N.M., D.O.O., D.M.P., A.R.P., A.J.R., I.R.S., P.V.R.S., B.D.S., P.M.T., G.J.W., T.A.W. and M.Y. are listed alphabetically. All authors contributed to and participated in the process and all were involved in writing and editing the manuscript.

Corresponding authors

Correspondence to James E. Herbert-Read or Ann Thornton .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Peer review

Peer review information.

Nature Ecology & Evolution thanks Camille Mellin, Prue Addison and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information.

Supplementary text and Tables 1 and 2.

Reporting Summary

Source data fig. 3.

Issue number, final rank and proportion heard of for each issue in round 1 and round 2.

Rights and permissions

Reprints and permissions

About this article

Cite this article.

Herbert-Read, J.E., Thornton, A., Amon, D.J. et al. A global horizon scan of issues impacting marine and coastal biodiversity conservation. Nat Ecol Evol 6 , 1262–1270 (2022). https://doi.org/10.1038/s41559-022-01812-0

Download citation

Received : 12 November 2021

Accepted : 24 May 2022

Published : 07 July 2022

Issue Date : September 2022

DOI : https://doi.org/10.1038/s41559-022-01812-0

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

This article is cited by

• J. B. Haugen
• A. L. Agnalt

npj Ocean Sustainability (2024)

Multiple ocean threats

Nature Ecology & Evolution (2023)

Late Cenozoic cooling restructured global marine plankton communities

• Adam Woodhouse
• Anshuman Swain
• Christopher M. Lowery

Nature (2023)

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Killer whales use specialized hunting techniques to catch marine mammals in the open ocean

Subpopulation of transient killer whales hunt sea lions, seals and whale calves off the coast of california.

Killer whales foraging in deep submarine canyons off the coast of California represent a distinct subpopulation that uses specialized hunting techniques to catch marine mammals, Josh McInnes at the University of British Columbia and colleagues report March 20 in the open-access journal PLOS ONE.

Killer whales ( Orcinus orca ) are found in oceans around the world, but they form separate populations, or 'ecotypes', that have their own social structure, food preferences and hunting behaviors. One ecotype, known as transient killer whales, specialize in hunting marine mammals. This ecotype can be divided into two groups -- inner coast whales that feed in shallow coastal waters, and outer coast whales that hunt in deep water -- but relatively little is known about the outer coast whales. Researchers compiled data from marine mammal surveys conducted between 2006 and 2018 and whale-watching ecotours between 2014 and 2021, to investigate the foraging behavior of outer coast transient killer whales around Monterey Submarine Canyon in California.

Members of this subpopulation were only sighted in open water and mainly preyed on California sea lions, grey whale calves and northern elephant seals. They use specialized techniques to hunt in open water, where prey can't easily be cornered. They often subdued their prey by ramming it with their head or body, and used their tail to hit or catapult sea lions into the air. The researchers identified two main types of foraging behavior -- distributed groups diving independently in the open water, and tightly coordinated groups foraging along the contours of the submarine canyons.

These results suggest that the outer coast whales are a distinct subpopulation that has developed specialized hunting techniques to catch marine mammals in this deep-water habitat. Their distinct foraging behaviors may be culturally transmitted from generation to generation, the authors say.

The authors add: "Transient (mammal-hunting) killer whales have been studied primarily in coastal shallow water habitats, and there is currently little known regarding their behavior in offshore and deep pelagic systems. This study highlights the complex foraging behavior and ecology of transients and how they act as apex predators in productive deep submarine canyon systems and how their behavior is linked to multiple marine mammal prey populations in the North Pacific Ocean."

• Dolphins and Whales
• Marine Biology
• Wild Animals
• Behavioral Science
• Right whale
• U.S. Navy Marine Mammal Program
• Hunting dog
• Baleen whale
• Bowhead Whale

Story Source:

Materials provided by PLOS . Note: Content may be edited for style and length.

Journal Reference :

• Josh D. McInnes, Kevin M. Lester, Lawrence M. Dill, Chelsea R. Mathieson, Peggy J. West-Stap, Stephanie L. Marcos, Andrew W. Trites. Foraging behaviour and ecology of transient killer whales within a deep submarine canyon system . PLOS ONE , 2024; 19 (3): e0299291 DOI: 10.1371/journal.pone.0299291

Cite This Page :

Explore More

• The Milky Way's Earliest Building Blocks
• Say Hello to Biodegradable Microplastics
• Bacteria and Human Colorectal Cancers
• Harnessing Hydrogen at Life's Origin
• How Thick Is the Ice On Europa?
• 'Odor Sensor' Male/ Female Blood Pressure
• Ancient Star Formed in Another Galaxy
• New Insights Into Early Human Migration
• Metamaterial: An Endless Domino Effect
• Quantum Tornado and Black Holes

Trending Topics

Strange & offbeat.