research paper in marine science

• 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?
• Calls for Papers
• 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.

Press Release

Study in nature: protecting the ocean delivers a comprehensive solution for climate, fishing and biodiversity.

Photograph by Southern Line Islands

Groundbreaking global study is the first to map ocean areas that, if strongly protected, would help solve climate, food and biodiversity crises.

London, UK (17 March 2021) —A new study published in the prestigious peer-reviewed scientific journal Nature today offers a combined solution to several of humanity’s most pressing challenges. It is the most comprehensive assessment to date of where strict ocean protection can contribute to a more abundant supply of healthy seafood and provide a cheap, natural solution to address climate change—in addition to protecting embattled species and habitats.

An international team of 26 authors identified specific areas that, if protected, would safeguard over 80% of the habitats for endangered marine species, and increase fishing catches by more than eight million metric tons. The study is also the first to quantify the potential release of carbon dioxide into the ocean from trawling, a widespread fishing practice—and finds that trawling is pumping hundreds of millions of tons of carbon dioxide into the ocean every year, a volume of emissions similar to those of aviation.

“Ocean life has been declining worldwide because of overfishing, habitat destruction and climate change. Yet only 7% of the ocean is currently under some kind of protection,” said Dr. Enric Sala, explorer in residence at the National Geographic Society and lead author of the study, Protecting the global ocean for biodiversity, food and climate .

“In this study, we’ve pioneered a new way to identify the places that—if strongly protected—will boost food production and safeguard marine life, all while reducing carbon emissions,” Dr. Sala said. “It’s clear that humanity and the economy will benefit from a healthier ocean. And we can realize those benefits quickly if countries work together to protect at least 30% of the ocean by 2030.”

To identify the priority areas, the authors—leading marine biologists, climate experts, and economists—analyzed the world’s unprotected ocean waters based on the degree to which they are threatened by human activities that can be reduced by marine protected areas (for example, overfishing and habitat destruction). They then developed an algorithm to identify those areas where protections would deliver the greatest benefits across the three complementary goals of biodiversity protection, seafood production and climate mitigation. They mapped these locations to create a practical “blueprint” that governments can use as they implement their commitments to protect nature.

The study does not provide a single map for ocean conservation, but it offers a first-in-kind framework for countries to decide which areas to protect depending on their national priorities. However, the analysis shows that 30% is the minimum amount of ocean that the world must protect in order to provide multiple benefits to humanity.

“There is no single best solution to save marine life and obtain these other benefits. The solution depends on what society—or a given country—cares about, and our study provides a new way to integrate these preferences and find effective conservation strategies,” said Dr. Juan S. Mayorga, a report co-author and a marine data scientist with the Environmental Market Solutions Lab at UC Santa Barbara and Pristine Seas at National Geographic Society.

The study comes ahead of the 15th Conference of the Parties to the United Nations Convention on Biological Diversity, which is expected to take place in Kunming, China in 2021. The meeting will bring together representatives of 190 countries to finalize an agreement to end the world’s biodiversity crisis. The goal of protecting 30% of the planet’s land and ocean by 2030 (the “30x30” target) is expected to be a pillar of the treaty. The study follows commitments by the United States, the United Kingdom, Canada, the European Commission and others to achieve this target on national and global scales.

Safeguarding Biodiversity

The report identifies highly diverse marine areas in which species and ecosystems face the greatest threats from human activities. Establishing marine protected areas (MPAs) with strict protection in those places would safeguard more than 80% of the ranges of endangered species, up from a current coverage of less than 2%.

The authors found that the priority locations are distributed throughout the ocean, with the vast majority of them contained within the 200-mile Exclusive Economic Zones of coastal nations.

The additional protection targets are located in the high seas—those waters governed by international law. These include the Mid-Atlantic Ridge (a massive underwater mountain range), the Mascarene Plateau in the Indian Ocean, the Nazca Ridge off the west coast of South America and the Southwest Indian Ridge, between Africa and Antarctica.

"Perhaps the most impressive and encouraging result is the enormous gain we can obtain for biodiversity conservation—if we carefully chose the location of strictly protected marine areas,” said Dr. David Mouillot, a report co-author and a professor at the Université de Montpellier in France. “One notable priority for conservation is Antarctica, which currently has little protection, but is projected to host many vulnerable species in a near future due to climate change."

Shoring up the Fishing Industry

The study finds that smartly placed marine protected areas (MPAs) that ban fishing would actually boost the production of fish—at a time when supplies of wild-caught fish are dwindling and demand is rising. In doing so, the study refutes a long-held view that ocean protection harms fisheries and opens up new opportunities to revive the industry just as it is suffering from a recession due to overfishing and the impacts of global warming.

“Some argue that closing areas to fishing hurts fishing interests. But the worst enemy of successful fisheries is overfishing—not protected areas,” Dr. Sala said.

The study finds that protecting the right places could increase the catch of seafood by over 8 million metric tons relative to business as usual.

“It’s simple: When overfishing and other damaging activities cease, marine life bounces back,” said Dr. Reniel Cabral, a report co-author and assistant researcher with the Bren School of Environmental Science & Management and Marine Science Institute at UC Santa Barbara. “After protections are put in place, the diversity and abundance of marine life increase over time, with measurable recovery occurring in as little as three years. Target species and large predators come back, and entire ecosystems are restored within MPAs. With time, the ocean can heal itself and again provide services to humankind.”

Soaking up Carbon

The study is the first to calculate the climate impacts of bottom trawling, a damaging fishing method used worldwide that drags heavy nets across the ocean floor. It finds that the amount of carbon dioxide released into the ocean from this practice is larger than most countries’ annual carbon emissions, and similar to annual carbon dioxide emissions from global aviation.

“The ocean floor is the world’s largest carbon storehouse. If we’re to succeed in stopping global warming, we must leave the carbon-rich seabed undisturbed. Yet every day, we are trawling the seafloor, depleting its biodiversity and mobilizing millennia-old carbon and thus exacerbating climate change. Our findings about the climate impacts of bottom trawling will make the activities on the ocean’s seabed hard to ignore in climate plans going forward,” said Dr. Trisha Atwood of Utah State University, a co-author of the paper.

The study finds that countries with the highest potential to contribute to climate change mitigation via protection of carbon stocks are those with large national waters and large industrial bottom trawl fisheries. It calculates that eliminating 90% of the present risk of carbon disturbance due to bottom trawling would require protecting only about 4% of the ocean , mostly within national waters.

Closing a Gap

The study’s range of findings helps to close a gap in our knowledge about the impacts of ocean conservation, which to date had been understudied relative to land-based conservation.

“The ocean covers 70% of the earth—yet, until now, its importance for solving the challenges of our time has been overlooked,” said Dr. Boris Worm, a study co-author and Killam Research Professor at Dalhousie University in Halifax, Nova Scotia. “Smart ocean protection will help to provide cheap natural climate solutions, make seafood more abundant and safeguard imperiled marine species—all at the same time. The benefits are clear. If we want to solve the three most pressing challenges of our century—biodiversity loss, climate change and food shortages —we must protect our ocean.”

Additional Quotes from Supporters and Report Co-Authors

Zac Goldsmith, British Minister for Pacific and the Environment, UK

Kristen Rechberger, Founder & CEO, Dynamic Planet

Dr. William Chueng, Canada Research Chair and Professor, The University of British Columbia, Principal Investigator, Changing Ocean Research Unit, The University of British Columbia

Dr. Jennifer McGowan, Global Science, The Nature Conservancy & Center for Biodiversity and Global Change, Yale University

Dr. Alan Friedlander, Chief Scientist, Pristine Seas, National Geographic Society at the Hawai'i Institute of Marine Biology, University of Hawai'i

Dr. Ben Halpern, Director of the National Center for Ecological Analysis and Synthesis (NCEAS), UCSB

Dr. Whitney Goodell, Marine Ecologist, Pristine Seas, National Geographic Society

Dr. Lance Morgan, President and CEO, Marine Conservation Institute

Dr. Darcy Bradley, Co-Director of the Ocean and Fisheries Program at the Environmental Market Solutions Lab, UCSB

The study, Protecting the global ocean for biodiversity, food and climate , answers the question of which places in the ocean should we protect for nature and people. The authors developed a novel framework to produce a global map of places that, if protected from fishing and other damaging activities, will produce multiple benefits to people: safeguarding marine life, boosting seafood production and reducing carbon emissions. Twenty-six scientists and economists contributed to the study.

Study’s Topline Facts

• Ocean life has been declining worldwide because of overfishing, habitat destruction and climate change. Yet only 7% of the ocean is currently under some kind of protection.
• A smart plan of ocean protection will contribute to more abundant seafood and provide a cheap, natural solution to help solve climate change, alongside economic benefits.
• Humanity and the economy would benefit from a healthier ocean. Quicker benefits occur when countries work together to protect at least 30% of the ocean.
• Substantial increases in ocean protection could achieve triple benefits, not only protecting biodiversity, but also boosting fisheries’ productivity and securing marine carbon stocks.

Study’s Topline Findings

• The study is the first to calculate that the practice of bottom trawling the ocean floor is responsible for one gigaton of carbon emissions on average annually. This is equivalent to all emissions from aviation worldwide. It is, furthermore, greater than the annual emissions of all countries except China, the U.S., India, Russia and Japan.
• The study reveals that protecting strategic ocean areas could produce an additional 8 million tons of seafood.
• The study reveals that protecting more of the ocean--as long as the protected areas are strategically located--would reap significant benefits for climate, food and biodiversity.

Priority Areas for Triple Wins

• If society were to value marine biodiversity and food provisioning equally, and established marine protected areas based on these two priorities, the best conservation strategy would protect 45% of the ocean, delivering 71% of the possible biodiversity benefits, 92% of the food provisioning benefits and 29% of the carbon benefits.
• If no value were assigned to biodiversity, protecting 29% of the ocean would secure 8.3 million tons of extra seafood and 27% of carbon benefits. It would also still secure 35% of biodiversity benefits.
• Global-scale prioritization helps focus attention and resources on places that yield the largest possible benefits.
• A globally coordinated expansion of marine protected areas (MPAs) could achieve 90% of the maximum possible biodiversity benefit with less than half as much area as a protection strategy based solely on national priorities.
• EEZs are areas of the global ocean within 200 nautical miles off the coast of maritime countries that claim sole rights to the resources found within them. ( Source )

Priority Areas for Climate

• Eliminating 90% of the present risk of carbon disturbance due to bottom trawling would require protecting 3.6% of the ocean, mostly within EEZs.
• Priority areas for carbon are where important carbon stocks coincide with high anthropogenic threats, including Europe’s Atlantic coastal areas and productive upwelling areas.

Countries with the highest potential to contribute to climate change mitigation via protection of carbon stocks are those with large EEZs and large industrial bottom trawl fisheries.

Priority Areas for Biodiversity

• Through protection of specific areas, the average protection of endangered species could be increased from 1.5% to 82% and critically endangered species from 1.1% to and 87%.
• the Antarctic Peninsula
• the Mid-Atlantic Ridge
• the Mascarene Plateau
• the Nazca Ridge
• the Southwest Indian Ridge
• Despite climate change, about 80% of today’s priority areas for biodiversity will still be essential in 2050. In the future, however, some cooler waters will be more important protection priorities, whereas warmer waters will likely be too stressed by climate change to shelter as much biodiversity as they currently do. Specifically, some temperate regions and parts of the Arctic would rank as higher priorities for biodiversity conservation by 2050, whereas large areas in the high seas between the tropics and areas in the Southern Hemisphere would decrease in priority.

Priority Areas for Food Provision

• If we only cared about increasing the supply of seafood, strategically placed MPAs covering 28% of the ocean could increase food provisioning by 8.3 million metric tons.

The Campaign for Nature works with scientists, Indigenous Peoples, and a growing coalition of over 100 conservation organizations around the world who are calling on policymakers to commit to clear and ambitious targets to be agreed upon at the 15th Conference of the Parties to the Convention on Biological Diversity in Kunming, China in 2021 to protect at least 30% of the planet by 2030 and working with Indigenous leaders to ensure full respect for Indigenous rights.

Media Contact

The National Geographic Society is a global nonprofit organization that uses the power of science, exploration, education and storytelling to illuminate and protect the wonder of our world. Since 1888, National Geographic has pushed the boundaries of exploration, investing in bold people and transformative ideas, providing more than 15,000 grants for work across all seven continents, reaching 3 million students each year through education offerings, and engaging audiences around the globe through signature experiences, stories and content. To learn more, visit www.nationalgeographic.org or follow us on Instagram , LinkedIn, and Facebook .

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

Evaluation of the Allen Coral Atlas benthic habitat map product for New Caledonia using representative habitat observations from a multi‑species fishery assessment

Serge, A., Maële, B., Stéphane, G. et al. Evaluation of the Allen Coral Atlas benthic habitat map product for New Caledonia using representative habitat observations from a multi-species fishery assessment. Coral Reefs (2024). https://doi.org/10.1007/s00338-024-02481-0.

Serge Andrefouet, a scientist who joined us on the Global Reef Expedition, assessed the accuracy of the Allen Coral Atlas's coral reef habitat maps in New Caledonia, and compared it to other reef maps, including the ground-truthed maps we created on the Global Reef Expedition, and found our maps to be more accurate.

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.

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

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

Articles on Marine science

Displaying 1 - 20 of 45 articles.

Through the magnifying glass: how cutting-edge technology is helping scientists understand baby corals

Marine Gouezo , Southern Cross University and Christopher Doropoulos , CSIRO

The Titan disaster investigation has begun. An expert explains what might happen next

Paola A. Magni , Murdoch University

The world’s fish are shrinking as the climate warms. We’re trying to figure out why

Timothy Clark , Deakin University

Dugongs and turtles are starving to death in Queensland seas – and La Niña’s floods are to blame

Kathy Ann Townsend , University of the Sunshine Coast

How you can help protect sharks – and what doesn’t work

David Shiffman , Arizona State University

Thousands of photos captured by everyday Australians reveal the secrets of our marine life as oceans warm

Gretta Pecl , University of Tasmania ; Barrett Wolfe , University of Tasmania ; Curtis Champion , Southern Cross University ; Jan Strugnell , James Cook University , and Sue-Ann Watson , James Cook University

Thousands of giant crabs amass off Australia’s coast. Scientists need your help to understand it

Elodie Camprasse , Deakin University

The ocean is not a quiet place

Thomas Uboldi , Université du Québec à Rimouski (UQAR)

Stinky seaweed is clogging Caribbean beaches – but a New Zealand solution could turn it into green power and fertiliser

Saeid Baroutian , University of Auckland, Waipapa Taumata Rau and Terrell Thompson , University of Auckland, Waipapa Taumata Rau

Australia’s marine industries deliver $80b a year. But without more scientists, the ‘blue economy’ is at risk

Toni Moate , CSIRO ; Anthony Boxshall , The University of Melbourne ; David Souter , Australian Institute of Marine Science , and Michelle Heupel , University of Tasmania

Snorkellers discover rare, giant 400-year -old coral – one of the oldest on the Great Barrier Reef

Adam Smith , James Cook University ; Nathan Cook , James Cook University , and Vicki Saylor , Indigenous Knowledge

Beautiful, rare ‘purple cauliflower’ coral off NSW coast may be extinct within 10 years

Meryl Larkin , Southern Cross University ; David Harasti , Southern Cross University ; Stephen D. A. Smith , Southern Cross University , and Tom R Davis

The Ocean Decade: how the next ten years can chart a new course for the blue planet

Jack Marley , The Conversation

In a first discovery of its kind, researchers have uncovered an ancient Aboriginal archaeological site preserved on the seabed

Jonathan Benjamin , Flinders University ; Geoff Bailey , University of York ; Jo McDonald , The University of Western Australia ; Michael O'Leary , The University of Western Australia , and Sean Ulm , James Cook University

Scientists have found oil from the Deepwater Horizon blowout in fishes’ livers and on the deep ocean floor

Steven Murawski , University of South Florida and Sherryl Gilbert , University of South Florida

A South African theatre project explores collective solutions to saving the ocean

Kira Erwin , Durban University of Technology

Iceland didn’t hunt any whales in 2019 – and public appetite for whale meat is fading

Joe Roman , University of Vermont

‘This situation brings me to despair’: two reef scientists share their climate grief

Jon Brodie , James Cook University and Alana Grech , James Cook University

We organised a conference for 570 people without using plastic. Here’s how it went

Elizabeth Sinclair , The University of Western Australia ; Dr Charlotte Birkmanis , The University of Western Australia , and Robert Pemberton , The University of Western Australia

Study identifies nine research priorities to better understand NZ’s vast marine area

Rebecca Jarvis , Auckland University of Technology and Tim Young , Auckland University of Technology

Related Topics

• Climate change
• Coral reefs
• Great Barrier Reef
• Marine biology
• Marine ecosystems
• Marine life
• Oceanography

Top contributors

Senior Curator of Marine Invertebrates, Museums Victoria Research Institute

Chair, National Marine Science Committee and Director, CSIRO National Collections and Marine Infrastructure, CSIRO

Chair professor, The University of Queensland

Distinguished Professor of Marine Science, King Abdullah University of Science and Technology

Professor, Griffith University

Professorial Fellow, ARC Centre of Excellence for Coral Reef Studies, James Cook University

Adjunct assistant professor, Southern Cross University

Senior Research Fellow, The University of Western Australia

Professor Marine Biology and Aquaculture, James Cook University

Director, Oceans and Atmosphere Flagship, CSIRO

Associate professor, Climate Change Research Centre, UNSW, UNSW Sydney

Associate professor, James Cook University

Professor of Coastal Oceanography, The University of Western Australia

Chief Executive Officer, Australian Institute of Marine Science

Senior Research Scientist, CSIRO

• X (Twitter)
• Unfollow topic Follow topic

The Pub Hub

Our research.

At the Perry Institute for Marine Science, we’re at the forefront of addressing critical challenges confronting our ocean. Through our cutting-edge conservation research, we delve into the intricate dynamics of fisheries, coral reefs, mangroves, and coastal habitats, all while spotlighting the profound interconnection between marine ecosystems and the livelihoods of coastal communities.

• Invertebrates

A Climate Vulnerability Assessment for Caribbean Recreational Fisheries (2022). Environmental Defense Fund , 37 pp.

Antillogorgia elisabethae Management Plan for The Commonwealth of The Bahamas (2021) . Prepared for the Department of Environmental Planning and Protection (DEPP) , Nassau, Bahamas. 77 pp

Contemporary and emerging fisheries in The Bahamas – conservation and management challenges, achievements and future direction (2018) . Fisheries Management and Ecology , doi: 10.1111/fme.12299.

Historical processes and contemporary anthropogenic activities influence genetic population dynamics of Nassau grouper (Epinephelus striatus) within The Bahamas (2017) . Frontiers in Marine Science , doi:10.3389/fmars.2017.00393.

Integrating Population Biology into Conservation Management for Endangered Nassau Grouper (Epinephelus striatus) (2016 ). Marine Ecology Progress Series, 554:263-280.

Multi-year tracking of Nassau grouper spawning migrations (2016) . Marine and Coastal Fisheries , 8:522–535

Nassau grouper migration patterns during full moon suggest collapsed historic fish spawning aggregation and evidence of an undocumented aggregation (2017) . Bulletin of Marine Science , 93:375–389

Nassau Grouper Policy Brief . Nassau, The Bahamas (2018), 7pp.

Nassau Grouper (Epinephelus striatus) Conservation Management Plan for The Commonwealth of The Bahamas . Prepared for the Department of Marine Resources , Nassau, Bahamas (2018).

RAD-seq analysis and in situ monitoring of Nassau grouper reveal fine-scale population structure and origins of aggregating fish (2020) . Frontiers in Marine Science , doi: 10.3389/fmars.2020.00157.

Review of the benefits of no-take zones . A report to the Wildlife Conservation Society 104 pp.

Spatial and temporal variability in parrotfish assemblages on Bahamian coral reefs (2020) . Diversity 14: 625.

Understanding and managing fish populations: keeping the toolbox fit for purpose (2018) . Journal of Fish Biology , doi:10.1111/jfb.13549.

Acknowledging and Supplanting White Supremacy Culture in Science Communication and STEM: The Role of Science Communication Trainers (2022). Frontiers in Communication, doi: 10.3389/fcomm.2022.787750.

Characterization of a thermally tolerant Orbicella faveolata reef in Abaco, The Bahamas . Coral Reefs , doi: 10.1007/s00338-020-01948-0.

Joining global efforts to halt coral reef decline: a call for more diversity, equity, and inclusion atinternational scientific meetings (2022). Reef Encounter: The new magazine of the international coral reef society.

Acropora cervicornis and Acropora palmata cultured on a low maintenance line nursery design in The Bahamas (2022). PLoS ONE, doi: 10.1371/journal.pone.0267034.

Bahamas Coral Reef Report Card Volume 2: 2015-2020

Bahamas Coral Reef Report Card (2016) , Volume 1: 2011-2013.

Characterization of a thermally tolerant Orbicella faveolata reef in Abaco, The Bahamas (2020) . Coral Reefs , doi: 10.1007/s00338-020-01948-0.

Census of heat tolerance among Florida’s threatened staghorn corals finds resilient individuals throughout existing nursery populations (2021).  Proceedings of the Royal Society B, doi: 10.1098/rspb.2021.1613. 

New Providence Stony Coral Tissue Loss Disease Preliminary Assessment. Nassau, Bahamas. Report submitted to the Department of Marine Resources . 9pp

New Providence and Rose Island, Bahamas Coral Reef Report Card (2014).

Rapid assessment of the occurrence of stony coral tissue loss disease (SCTLD) along the southern coast of Grand Bahama, Bahamas (2020).

Tourism and coral reef health in the Exuma Cays Land and Sea Park. Proceedings of the 65th Gulf and Caribbean Fisheries Institute (2013).

An assessment of the queen conch resources of the East Grand Bahama National Park (2019). 9pp

Bahamas Coral Reef Report Card , Volume 1: 2011-2013 (2016).

Bahamas protected: Realizing the 2020 goal to effectively manage and expand Bahamian marine protected areas. A report prepared for the Ministry of the Environment for The Bahamas on the ecological gap analysis workshop held in Nassau, New Providence (2016). The Nature Conservancy, Bahamas National Trust, and Bahamas Reef Environment Educational Foundation, Nassau, Bahamas. 41 pp

Expanding The Bahamas marine protected area network to protect 20% of the marine and coastal environment by 2020: a gap analysis (2017). 66pp.

Marine biophysical monitoring plan for The Bahamas National Protected Area System (BNPAS) (2020) . Submitted to the Bahamas National Trust , Nassau, Bahamas.

Monitoring program for The Bahamas National Protected Area System. Methods for assessing the ecological condition of coral reefs, seagrasses and mangroves (2014). Report submitted to the BEST Commission under the GEF FSP Pilot 3 Demonstration Project “Tourism and coral reef health in the Exuma Cays Land and Sea Park ”. 22 pp.

Preliminary assessment of Hurricane Dorian’s impacts on coral reefs of Abaco and Grand Bahama. Report submitted to the Government of The Bahamas. 28pp

Rapid ecological assessment for the expansion of Lucayan National Park (2014). Rapid ecological assessment for the expansion of Lucayan National Park . 34pp.

20 by 20 White paper: Marine protection plan for expanding The Bahamas Marine Protected Areas Network to meet The Bahamas 2020 declaration (2018) 137 pp.

• Work With Us
• Reef Rescue Network

Join Our Coralition

Subscribe to our newsletter to stay updated on our latest research missions, captivating stories, and exciting job opportunities.

• Perry Institute for Marine Science

Leading through science, we innovate solutions, create opportunities, and inspire action to protect and restore ocean life for people and the planet.

Read Our Stories

Celebrating earth day and mangrove restoration in grand bahama, urgent: train bahamian youth to restore coral, spawning aggregation fisheries workshop.

• Report Stony Coral Tissue Loss Disease

Where We Are

Windsor School (Albany Campus) Frank Watson Boulevard Adelaide, The Bahamas

• 5356 Main Street, Suite 2
• P.O. Box 435
• Waitsfield, VT 05673

EDITORIAL article

This article is part of the research topic.

Deep Learning for Marine Science

Editorial: Deep Learning for Marine Science Provisionally Accepted

• 1 Ocean University of China, China
• 2 Center for Environmental Science, University of Maryland, College Park, United States
• 3 Shenzhen International Graduate School, Tsinghua University, China
• 4 Louisiana State University, United States

The final, formatted version of the article will be published soon.

In recent years, Deep Learning (DL) technology has been widely used in marine science and technology research, and provides powerful technical support for related researches and applications. As ocean observation technology continues to advance, the volume of data generated by marine scientific research is steadily increasing. This offers vast potential for data-driven DL to demonstrate its capabilities and has therefore emerged as a valuable technology across multiple research fields, including biology, ecosystems, climate, energy, as well as physical and chemical interactions. The research topic “Deep Learning for Marine Science” aims to provide a research collection to collect relevant research work on the application of DL technology in marine science. A total of 39 papers are published with contributions by 236 authors. The contents in these papers focus on the following aspects: research survey, marine/underwater image enhancement/restoration/compression, marine/underwater visual recognition/detection, dataset and labeling, marine process/phenomenon prediction/detection, marine physical/biogeochemical variable prediction/reconstruction, and marine optics/acoustics. Here, we summarize the contents of these papers and highlight their key contributions to the research topic.

Keywords: Research survey, marine/underwater image enhancement/restoration/compression, marine/underwater visual recognition/detection, dataset and labeling, marine process/phenomenon prediction/detection, marine physical/biogeochemical variable prediction/reconstruction, marine optics/acoustics

Received: 26 Mar 2024; Accepted: 23 Apr 2024.

Copyright: © 2024 Zheng, Bi, Cheng and Benfield. 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) or licensor 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: Prof. Haiyong Zheng, Ocean University of China, Qingdao, China

People also looked at

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
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 15 February 2018

Ocean science research is key for a sustainable future

• Martin Visbeck 1 , 2  

Nature Communications volume  9 , Article number:  690 ( 2018 ) Cite this article

36k Accesses

200 Citations

247 Altmetric

Metrics details

• Ocean sciences
• Sustainability

Human activity has already affected all parts of the ocean, with pollution increasing and fish-stocks plummeting. The UN’s recent announcement of a Decade of Ocean Science provides a glimmer of hope, but scientists will need to work closely with decision-makers and society at large to get the ocean back on track.

The ocean covers 71% of the Earth’s surface. It regulates our climate and holds vast and in some cases untouched resources. It provides us with basics such as food, materials, energy, and transportation, and we also enjoy the seascape for religious or recreational practices. Today, more than 40% of the global population lives in areas within 200 km of the ocean and 12 out of 15 mega cities are coastal. Doubling of the world population over the last 50 years, rapid industrial development, and growing human affluence are exerting increasing pressure on the ocean. Climate change, non-sustainable resource extraction, land-based pollution, and habitat degradation are threatening the productivity and health of the ocean (Fig.  1 ). It is in this context that over the last few years, scientists and societal actors have organized a bottom-up movement, which has ultimately led to the United Nations General Assembly proclaiming a Decade of Ocean Science for Sustainable Development (2021–2030). In the process, governments, industry, and scientists have raised awareness of the rapid degradation and over-use of the ocean. The final document from the Rio+20 summit, The future we want 1 , made extensive reference to the ocean, and the Global Ocean Commission articulated the need for more effective global ocean policies 2 . Moreover, the 2030 Agenda for Sustainable Development includes an explicit ocean goal (SDG14) 3 , 4 that led to the first-ever UN Ocean conference 5 to support its implementation. The ambition of the Decade of Ocean Science is to now use this gathering momentum to mobilize the scientific community, policy-makers, business, and civil society around a program of joint research and technological innovation 6 . I see reasons for optimism in four main areas. First, there is a tremendous opportunity to connect ocean sciences more directly with societal actors by promoting integrated ocean observation and solution-oriented research agendas (Fig.  2 ). Also, rich and poor nations are increasingly engaging in capacity development and resource sharing. And finally, the UN system and coastal states have a unique chance to seriously collaborate in multi-stakeholder processes to advance maritime spatial planning and effective ocean governance.

Human–ocean interactions highlighting ocean ecosystem services and their threads (taken from Ocean Atlas, 2017 12 )

Increased need for ocean information to meet societal needs. Meeting a growing range of societal demands and achieving Sustainable Development Goal 14 (Conserve and sustainably use the oceans, seas, and marine resources) from the 2030 Agenda for Sustainable Development can only be fully realized if all elements of the ocean value chain are resourced adequately and more integrated science agendas are advanced. Figure adapted from ref. 9 . Sustainable Development Goal logo ©UNITED NATIONS. All rights reserved

A more integrated and sustainable ocean observing system

The Decade of Ocean Science will encourage actions towards a more integrated and sustainable ocean observing system to facilitate ocean discovery and environmental monitoring. The vast volume of the ocean and its complex coastlines are neither fully observed nor adequately understood. In particular the deep sea is a frontier of ocean sciences, where internationally coordinated research teams regularly discover new ocean phenomena of profound importance or new organisms and substances for potential future use. Sustained and systematic ocean observations are needed to document ocean change, initialize ocean system models and provide critical information for improved ocean understanding. Advances in ocean robotics and the combination of remote and in situ ocean observations offer new opportunities; and free and open data sharing and multi-stakeholder contributions by governments (rich and poor), the private sector and citizens are opening exciting new dimensions. International efforts, such as the Global Ocean Observing System, the Blue Planet initiative of the Group on Earth Observations and their Framework for Ocean Observing 7 provide a solid basis and an opportunity for growth. The upcoming decadal conference on ocean observations, OceanObs 19 8 will provide an excellent opportunity to advance our ocean observing ambitions.

A solution-oriented integrated ocean science agenda

The Decade of Ocean Science should address both deep disciplinary understanding of ocean processes and solution-oriented research to generate new knowledge. This will support societal actors in reducing ocean pressures, preserving and restoring ocean ecosystems and so safeguard ocean-related prosperity for generations to come. A solution-oriented integrated ocean science agenda can provide innovative ideas, improved assessments and fundamental knowledge in the context of sustainable development and improving human–ocean interactions 9 .

Our rapidly growing, affluent, and more technologically advanced societies are increasingly impacting their local and the global environment, leading to pollution by both chemical and physical wastes. Integrated research is needed to assess the human and environmental risks of ongoing and future types of ocean pollution, to generate new ideas to reduce the ocean pressures by promoting recycling, improved waste management and incentive and governance regimes to encourage more sustainable production and consumption. The most challenging ocean pollutants include: atmospheric carbon dioxide, which causes climate change, ocean warming, ocean acidification, and sea level rise; agricultural fertilizers, which lead to increased primary production but result in ocean deoxygenation; untreated waste water; invasive species; micro and macro plastics, the exponential increase of which has an environmental impact as yet only partially known.

Ocean hazards such as storm surges, harmful algal blooms, or coastline erosion can be devastating for coastal communities. Throughout human evolution civilization has developed strategies to increase our resilience to threats from the ocean. However, the rush for coastal recreation and access to the sea has produced newly built infrastructure that is increasingly vulnerable to ocean extreme events. Hard solutions, such as building walls and levees, could provide some mitigation. However, softer approaches, such as beach nourishment, restoration of mangroves and reef systems, would also provide natural protection and increase resilience to sea-level rise and storm surges. Marine protected areas, natural coastal defences, mining codes, or regulations to limit ocean pollution are all critical elements to safeguard ocean resilience.

Humans have always benefitted from the ocean and its diverse ecosystem services. We often speak of a healthy and productive ocean referring to the desire to maintain the ocean in a prosperous state. The ocean provides food for many, often poor, coastal communities; provides jobs, energy, and raw materials; and enables global trade and recreational and cultural services. The sustainability challenge is achieving long-term ocean prosperity for more affluent societies with a global population approaching 10 billion. Is there sufficient intergenerational will to sustain the overall long-term wealth and well-being of humans by safeguarding ocean resources and ecosystem productivity? What are the trade-offs and synergies between different strategies of marine food production and wild harvesting, different forms of energy harvesting and extraction of materials and ocean restoring zones? New research should develop and flesh out sustainable blue-green growth agendas and link it to efforts in ecosystem protection.

The Decade of Ocean Science should develop a new ocean narrative that can provide context and motivation to reduce ocean pressures, increase ocean resilience, and promote ocean prosperity for generations to come. At the same time, it can provide visibility to existing and new international ocean science programs, such as the new Future Earth Ocean Knowledge–Action network 10 that aims to connect academic and practical knowledge to address the pressing issues of ocean sustainability using the concept of co-design, co-production, and co-dissemination of ocean sustainability knowledge.

Global capacity building

The success of the Decade of Ocean Science will critically depend on global capacity building and resource-sharing between countries at different levels of wealth and development. The enormous need for more ocean information at the scientific, governmental, private sector, and public levels demands a step-change in ocean education at all levels. New technology to improve ocean observation, more sustainable ocean resource extraction, and the digital revolution are transforming the ocean sciences and information communities. How can we harness this opportunity? Perhaps new curricula at universities can provide the opportunity to engage a wide range of disciplines in the area of ocean sustainability. Global learning formats such as massive open online courses 11 , open access to ocean information and increased interactions between the academic and societal actor communities are all promising activities. In addition, partnerships between academic and civil society organizations can produce free ocean literacy material, such as the Ocean-Atlas 12 or the World Ocean Review 13 . However, more engagement at the primary and secondary school levels is urgently needed to promote ocean literacy. Training courses and exchange programs between south–south and north–south ocean actors, as well as courses for ocean professionals, hold tremendous potential to raise ocean awareness and promote better solutions.

Effective ocean governance

Finally, the Decade of Ocean Science, in conjunction with the 2030 Agenda for Sustainable Development and other international and regional ocean agendas, has shown the need for societal actors to reflect on effective ocean governance. From a regulatory perspective, coastal states can benefit from a systematic, multi-stakeholder assessment and spatial planning procedure. In many parts of the world, each cubic meter of ocean is expected to support several, often conflicting, demands. Spatial planning procedures that take the demands of neighboring countries and the global ocean system into account can help to find more sustainable and equitable regimes of ocean use and access. Science can help in this effort by reflecting on a range of human development scenarios and evaluating how best to sustain ocean prosperity while respecting planetary 14 and ocean boundaries. A good example of this is the ocean scenario team that is scoping out development pathways to reach SDG14 and ocean-related goals 15 by 2050 in the context of The World In 2050 project (TWI2050 16 ).

The increased awareness of the importance of the ocean to the future of humanity give grounds for cautious optimism and motivation for ambitious multilateral cooperation. The scientific community has been given a stage on which to shine during the Decade of Ocean Science for Sustainable Development. Let us come together, respect our disciplines and agendas but also be ready to embark on an exciting and transformative journey to realize the ocean we need for the future we want.

United Nations General Assembly. The future we want . Rio+20 conference outcome document A/RES/66/288. Available at https://sustainabledevelopment.un.org/futurewewant.html (United Nations, 2012).

Global Ocean Commission. The Future of Our Ocean: Next steps and priorities Report available at http://www.some.ox.ac.uk/research/global-ocean-commission (Global Ocean Commission, 2016).

Visbeck, M. et al. Securing blue wealth: the need for a special sustainable development goal for the ocean and coasts and for future ocean spatial planning. Mar. Policy 48 , 184–191 (2014).

Article   Google Scholar  

United Nations General Assembly. Transforming our world: the 2030 Agenda for Sustainable Development. A/RES/70/1 Available at http://www.un.org/sustainabledevelopment/development-agenda/ (United Nations, 2015).

United Nations. UN Ocean Conference 2017 . https://oceanconference.un.org/ (2017).

UNESCO. United Nations Decade of Ocean Science for Sustainable Development (2021-2030) UNESCO press release. https://en.unesco.org/ocean-decade (2017).

Task Team for an Integrated Framework for Sustained Ocean Observing.  A Framework for Ocean Observing.  IOC/INF-1284 (UNESCO, 2012).

OceanObs'19. OceanObs'19 Conference Announcement.   https://www.oceanobs19.net (2018).

Visbeck, M. & Schneider, A. Future Earth’s Ocean in Global Change and Future Earth: The GeoScience Perspective  (eds Beer, T.,  Li, J. &  Alverson, K.) in press (Cambridge University Press, 2018).

Ocean-KAN development team. Future Earth Ocean Knowledge Action Network .  http://www.futureearth.org/future-earth-ocean (2018).

Antia, A. & Visbeck, M. (eds) One Planet – One Ocean: From Science To Solutions, Massive Open Online Course . Available at  https://www.oceanmooc.org (2017).

Visbeck, M. et al.  Ocean-Atlas, Facts and Figures on the Threats to Our Marine Ecosystems. https://www.ocean-atlas.org (2017).

Lehmköster, J. (ed.) The World Ocean Review Series   http://worldoceanreview.com/en/ (2010, 2013, 2014, 2015 & 2017).

Steffen, W. et al. Planetary boundaries: Guiding human development on a changing planet.  Science 347 , 1259855 (2015).

Article   PubMed   Google Scholar  

Schmidt S. et al. A Guide to SDG Interactions: from Science to Implementation. Ch.SDG14. Report available at  https://www.icsu.org/publications/a-guide-to-sdg-interactions-from-science-to-implementation (2017).

TWI2050 project team.  The World in 2050 .  http://www.iiasa.ac.at/web/home/research/researchProjects/TWI2050.html (2017).

Download references

Author information

Authors and affiliations.

GEOMAR Helmholtz Centre for Ocean Research Kiel, Wischhofstr. 1-3, 24118, Kiel, Germany

Martin Visbeck

Kiel University, Christian-Albrechts-Platz 4, 24118, Kiel, Germany

You can also search for this author in PubMed   Google Scholar

Contributions

M.V. wrote the manuscript.

Corresponding author

Correspondence to Martin Visbeck .

Ethics declarations

Competing interests.

The authors declare no competing financial interests.

Additional information

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

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Visbeck, M. Ocean science research is key for a sustainable future. Nat Commun 9 , 690 (2018). https://doi.org/10.1038/s41467-018-03158-3

Download citation

Received : 15 January 2018

Accepted : 24 January 2018

Published : 15 February 2018

DOI : https://doi.org/10.1038/s41467-018-03158-3

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

In search of value: the intricate impacts of benefit perception, knowledge, and emotion about climate change on marine protection support.

• Minh-Hoang Nguyen
• Minh-Phuong Thi Duong
• Vuong-Quan Hoang

Journal of Environmental Studies and Sciences (2024)

Principles for transformative ocean governance

• Amanda T. Lombard
• Jai Clifford-Holmes
• Elisa Morgera

Nature Sustainability (2023)

Isolation and screening of antimicrobial biosurfactants obtained from mangrove plant root–associated bacteria

• Zainab Irfan
• Sayeed Mohammed Firdous
• Eswaramoorthy Thirumalaikumar

Naunyn-Schmiedeberg's Archives of Pharmacology (2023)

A critical review of conventional and emerging wastewater treatment technologies

• Ravindra Sangamnere
• Tavishi Misra
• Rakesh Kumar

Sustainable Water Resources Management (2023)

Understanding and Adapting Ocean Decade Action

• Dhinakarasamy Inbakandan

Anthropocene Science (2023)

Quick links

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

Sign up for the Nature Briefing: Anthropocene newsletter — what matters in anthropocene research, free to your inbox weekly.

Meet the 2024 Knauss Fellows making waves in ocean science for sustainable development

• April 30, 2024

Ten Years. Ten Challenges. One Ocean.

The UN Decade of Ocean Science for Sustainable Development is the largest global ocean science initiative ever taken. Taking place from 2021-2030, this decade-long framework seeks to stimulate ocean science and knowledge generation to reverse the decline of the state of the ocean, and catalyze new opportunities for a sustainable marine economy. 

From April 12-14, 2024, over 1,500 scientists, government and UN officials, philanthropic organizations, industry leaders, students, and communicators came together to celebrate recent accomplishments and look towards the future of scientific contributions at the 2024 Ocean Decade Conference . 

Two of NOAA Research’s Knauss Fellows were on the ground in Barcelona, Spain. Dr. Jai Tiarks (they/them) has a background in freshwater ecology and is a Knauss Fellow in NOAA Research’s Office of International Activities. Dr. Emily Nocito (she/her) is an Ocean Decade Knauss Fellow whose background is in international marine governance and conservation.

We were able to sit down with Jai and Emily to learn more about their work in ocean science and their experiences supporting this global initiative for ocean health. Continue reading to hear what they had to say.

Tell me a little bit about your role at NOAA Research. 

Jai:  My Knauss Fellowship in the Office of International Activities has given me the opportunity to build a skillset in international collaboration. Along with my office, I help facilitate collaboration with NOAA labs and programs and international partners.

Emily: I’m really lucky in that my Knauss Fellowship in NOAA Research allows me to work in two different but important areas. One area is related to the domestic implementation of the UN Ocean Decade – in this role, I serve as the USA Early Careers Ocean Professional (ECOP) Node coordinator. As the USA ECOP Node coordinator, I engage and communicate with our members and other relevant organizations, and work with other national and regional coordinators and the Global ECOP Programme. I also work closely with the US National Committee for the Ocean Decade , which is hosted by the National Academies of Science, and I work closely with other Federal stakeholders as the Executive Secretary for both the Task Force on Ocean Decade and the Interagency Working Group on Ocean Decade. My Knauss Fellowship also has me working with the NOAA Science Council , where I sit as one of the executive secretaries.

What inspired you to work in ocean science?

Jai: I’m actually relatively new to marine sciences. While my background is in freshwater systems, I’m drawn to the vastness of the ocean. Oceans cover about 70% of the planet and serve as a point of connection between many countries. Because of this interconnectedness, oceans (arguably) require collaboration and coordinated efforts that scale from the community level all the way to the intergovernmental level in order to ensure healthy oceans. 

Emily:  I love complex issues – my background is in ocean governance and conservation decision-making. To me, the ocean is one of the most complex spaces that humanity is involved in. It is largely unexplored; it is a global common; it is a place of beauty and wonder, political strife and environmental crises. The dichotomy of the ocean means that every day I am intrigued by my work.

While NOAA’s involvement is focused on the science of the UN Decade of Ocean Science for Sustainable Development, what societal impacts do you hope to see achieved by the end of 2030?

Jai: I believe science that is actionable and applicable ends up being the most impactful. I think the latter half of the UN Ocean Decade name – science for sustainable development – really emphasizes the need to prioritize applicable science; science that can actually inform society in a tangible way. My hope is that, by focusing on all the ways we are intertwined with the ocean, we can make positive and meaningful changes toward a healthy and sustainable ocean.

Emily: Science for the sake of science is fine, but it is through tangible and purposeful science that we see positive impacts on society and development. There will always be a humanitarian aspect to science. The “call to action” for the Ocean Decade is “the science we need for the ocean we want”, and the Ocean Decade for Sustainable Development has 10 challenges we need to overcome to reach the 7 Decade Outcomes . Actions all work towards overcoming these challenges to achieve the outcomes. The cool part about this is, anyone can submit an Action , as long as it addresses one of the 10 challenges of the UN Ocean Decade for Sustainable Development. I hope that the science we produce will lead to a safe, equitable, accessible and healthy ocean for generations to come.

Learn more about the role of NOAA in the Ocean Decade here .

Meet John Cortinas, NOAA Research’s New Deputy Assistant Administrator for Science

Meet the women of NOAA advancing our knowledge of a changing Arctic 

Meet the women advancing NOAA’s exploration and stewardship of the ocean and Great Lakes 

Meet the women advancing NOAA’s marine heatwave research

Popup call to action.

A prompt with more information on your call to action.

share this!

April 29, 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

Climate change reveals intricate dynamics of reproductive barriers in marine species

by Monash University

Monash University scientists have uncovered insights into how rising temperatures influence the reproductive interactions and species boundaries of marine organisms.

As climate change continues to reshape ecosystems, a new study from the Faculty of Science at Monash University sheds light on the complex interplay between temperature , parental sex and reproductive barriers in marine tubeworms, known as Galeolaria, from southern Australia's fast-warming biodiversity hotspot.

Galeolaria tubeworms are foundation species found on the rocky shores of temperate Australia. The tubeworms build and occupy dense colonies of stony tubes, which enhance coastal biodiversity by providing habitat and refuge from heat stress for species that would not otherwise persist there.

The study , published in Evolution , was motivated by the fact that climate change is altering species ranges and interactions, providing species with new opportunities to mate and hybridize.

The research team examined a range of barriers to reproduction across the life stages of two Galeolaria species, spanning different temperature ranges from Cape Otway in Victoria all the way up to Tathra in New South Wales. This region is globally recognized for its marine biodiversity and is also warming much faster than the global average rate.

The findings indicate that reproductive barriers between the species are weaker at the fertilization and embryogenesis stage, but become stronger and more temperature-sensitive during the larval development stage.

Importantly, the study also reveals asymmetry in barriers between parental sexes, suggesting a complex interplay between thermal niche differentiation and maternal inheritance.

Senior author Dr. Keyne Monro from Monash University School of Biological Sciences explains, "Our research highlights the pivotal role of temperature in shaping reproductive isolation between species of Galeolaria. As temperatures rise, we observe shifts in the strength of reproductive barriers, with significant implications for the future dynamics of species interactions and biodiversity.

"Our findings point to a key role for temperature in reproductive isolation, but also challenges for predicting the fate of isolation in future climates."

The study underscores the importance of understanding the environmental and biological factors that influence reproductive isolation, particularly in the context of climate change.

By examining barriers across multiple life stages and environmental conditions , researchers can gain valuable insights into the mechanisms driving species sensitivities to climate and species hybridization.

"Our findings emphasize the need for comprehensive assessments of reproductive barriers across diverse environments," says co-author Associate Professor Kay Hodgins from the Monash University School of Biological Sciences.

"This knowledge is crucial for predicting the fate of species boundaries and guiding conservation efforts in a rapidly changing world.

"By unraveling the intricate relationship between temperature, reproductive barriers, and species interactions, scientists are better equipped to protect and preserve marine biodiversity in the face of environmental change," said Dr. Hodgins.

As marine ecosystems face unprecedented challenges from climate change, studies like this provide essential insights into the adaptive strategies and ecological dynamics of marine organisms.

Journal information: Evolution

Provided by Monash University

Explore further

Feedback to editors

New observatory in Chile—the highest in the world—aims to reveal origins of planets, galaxies and more

11 hours ago

Study uncovers the secret of long-lived stem cells

14 hours ago

Scientists show that ancient village adapted to drought, rising seas

15 hours ago

How polyps of the moon jellyfish repel viral attacks on their microbiome

Organic electrochemical transistors: Scientists solve chemical mystery at the interface of biology and technology

16 hours ago

Activity in a room stirs up nanoparticles left over from consumer sprays, study shows

Study dispels myth that purebred dogs are more prone to health problems

17 hours ago

Study shows climate change and mercury pollution stressed plants for millions of years

18 hours ago

Exploiting disorder to harvest heat energy: The potentialities of 2D magnets for thermoelectric applications

Citizen scientists help discover record-breaking exoplanet in binary star system

Relevant physicsforums posts, the cass report (uk).

13 hours ago

Is 5 milliamps at 240 volts dangerous?

Apr 29, 2024

Major Evolution in Action

Apr 22, 2024

If theres a 15% probability each month of getting a woman pregnant...

Apr 19, 2024

Can four legged animals drink from beneath their feet?

Apr 15, 2024

Mold in Plastic Water Bottles? What does it eat?

Apr 14, 2024

More from Biology and Medical

Related Stories

Global warming and invasive species threaten seagrass meadows in the Mediterranean

Feb 6, 2024

How do temperature extremes influence the distribution of species?

Nov 14, 2023

Declining male fertility increases climate change vulnerability

Apr 20, 2021

Climate change disrupts core habitats of marine species, finds modeling study

Feb 15, 2023

Climate change challenges marine conservation efforts in Atlantic Canada

Oct 4, 2023

Prezygotic isolation between sympatric dioecious figs may not yet be established

Oct 13, 2023

Recommended for you

Mimicry allows lesser necklaced laughingthrush birds to benefit from living among larger related species

19 hours ago

Missing link in species conservation: Pharmacists, chemists could turn tide on plant, animal extinction

20 hours ago

Researchers find pesticides in a third of Australian frogs tested. Did these cause mass deaths?

Adaptation of photosynthetic mechanism in air plants occurs through gene duplication, study finds

21 hours ago

Study is first to describe and recreate lemurs' one-of-a-kind vocal structure

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

Curiosity Rover Science

Landing at Gale Crater, Mars Science Laboratory is assessing whether Mars ever had an environment capable of supporting microbial life. Determining past habitability on Mars gives NASA and the scientific community a better understanding of whether life could have existed on the Red Planet and, if it could have existed, an idea of where to look for it in the future.

Science Objectives

To contribute to the four Mars exploration science goals and meet its specific goal of determining Mars' habitability, Curiosity has the following science objectives:

Biological objectives

Geological and geochemical objectives, planetary process objectives, surface radiation objective.

1. Determine the nature and inventory of organic carbon compounds 2. Inventory the chemical building blocks of life (carbon, hydrogen, nitrogen, oxygen, phosphorous, and sulfur) 3. Identify features that may represent the effects of biological processes

1. Investigate the chemical, isotopic, and mineralogical composition of the Martian surface and near-surface geological materials 2. Interpret the processes that have formed and modified rocks and soils

1. Assess long-timescale (i.e., 4-billion-year) atmospheric evolution processes 2. Determine present state, distribution, and cycling of water and carbon dioxide

Characterize the broad spectrum of surface radiation, including galactic cosmic radiation, solar proton events, and secondary neutrons

Science Highlights

With over a decade of exploration, Curiosity has unveiled the keys to some of science's most unanswered questions about Mars. Did Mars ever have the right environmental conditions to support small life forms called microbes? Early in its mission, Curiosity's scientific tools found chemical and mineral evidence of past habitable environments on Mars. It continues to explore the rock record from a time when Mars could have been home to microbial life.

Science Instruments

From cameras to environmental and atmospheric sensors, the Curiosity rover has a suite of state-of-the-art science instruments to achieve its goals.

Discover More Topics From NASA

James Webb Space Telescope

Perseverance Rover

Parker Solar Probe

Help | Advanced Search

Computer Science > Computation and Language

Title: phi-3 technical report: a highly capable language model locally on your phone.

Abstract: We introduce phi-3-mini, a 3.8 billion parameter language model trained on 3.3 trillion tokens, whose overall performance, as measured by both academic benchmarks and internal testing, rivals that of models such as Mixtral 8x7B and GPT-3.5 (e.g., phi-3-mini achieves 69% on MMLU and 8.38 on MT-bench), despite being small enough to be deployed on a phone. The innovation lies entirely in our dataset for training, a scaled-up version of the one used for phi-2, composed of heavily filtered web data and synthetic data. The model is also further aligned for robustness, safety, and chat format. We also provide some initial parameter-scaling results with a 7B and 14B models trained for 4.8T tokens, called phi-3-small and phi-3-medium, both significantly more capable than phi-3-mini (e.g., respectively 75% and 78% on MMLU, and 8.7 and 8.9 on MT-bench).

Submission history

Access paper:.

• 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 .