Submerged kelp forests are vital coastal ecosystems that support marine biodiversity and ecosystem dynamics, yet accurate underwater kelp segmentation remains challenging due to optical degradation, illumination variability, turbidity, overlapping vegetation, and complex benthic backgrounds. We systematically evaluated three deep learning semantic segmentation frameworks, ResNet34-U-Net, ResNet50-DeepLabV3, and a hybrid ResNet50-ASPP-Transformer architecture, for kelp detection using high-resolution underwater RGB imagery collected from northeastern U.S. coastal waters. A dataset of 3,395 SSeg assisted annotated image-mask pairs was developed for model training and validation, while geographically independent sites were used for quantitative and qualitative evaluation. All models used consistent preprocessing, augmentation, and evaluation protocols. On independent test data, ResNet50-DeepLabV3 achieved the highest Dice (0.7120) and Intersection over Union (IoU; 0.6267), followed by ResNet34 U Net (Dice 0.6868; IoU 0.5978). The hybrid ASPP Transformer achieved the highest pixel accuracy (0.8528) but lower Dice (0.6437) and IoU (0.5746). External qualitative evaluation further showed that DeepLabV3 produced more consistent segmentation across varying environmental conditions, image qualities, and benthic habitats. Overall, ResNet50-DeepLabV3, termed Kelp-O-Tron, provided the best balance of segmentation accuracy, robustness, and generalization. The dataset, annotation workflow, and comparative evaluation provide resources for advancing automated underwater habitat mapping and ecological monitoring.

Coral reefs are experiencing unprecedented change due to climate-driven disturbances, yet effective monitoring remains constrained by limited integration across data types and spatial scales. This study presents a multi-source framework for detecting structural and compositional changes in coral reefs by integrating ICESat-2-derived terrain metrics, high-resolution PlanetScope imagery, and field-based photoquadrat observations. We applied this workflow to Heron Reef in the southern Great Barrier Reef, Australia, to quantify changes between 2020 and 2025-a period marked by 3 mass bleaching events and 9 cyclones. ICESat-2 data provided rugosity, slope, and bathymetric depth, while spectral unmixing of PlanetScope imagery yielded subpixel estimates of benthic composition. Satellite derived bathymetry (SDB) was validated against ICESat-2 depths using a random forest regression (RMSE: 0.30-0.32 m). Temporal changes were quantified per pixel, filtered by uncertainty thresholds from the Special Law of Propagation of Variances (SLOPOV), and analyzed for spatial clustering using Moran's I. Results revealed structural and compositional shifts concentrated along the reef crest, with increases in rugosity and slope indicating potential recovery in some areas, while decreases suggested localized degradation. Coral cover declined modestly (55.4% to 53.8%), though spatial patterns were highly heterogeneous. While this framework enables broad-scale monitoring, it operates within the inherent uncertainties of large-area remote sensing, emphasizing the need for further in situ validation and application across diverse reef systems. This integrated workflow demonstrates a scalable, repeatable approach for comprehensive reef monitoring that leverages openly accessible satellite data and provides actionable insights for conservation management.

Kelps form ecologically important habitats around the globe but are threatened by anthropogenic stressors in much of their range. Within the Gulf of Maine, these stressors include rising ocean temperatures and species invasions. Monitoring these habitats is important, but our ability to do so varies regionally based on kelp species. Modelling techniques based on optical satellite imagery are useful for floating kelps but can only identify the subsurface kelps found in the Gulf of Maine within a small upper portion of their depth range. We developed an integrative approach to kelp habitat classification using two existing data sources: sea surface temperature data from Landsat 8 and high-resolution acoustic bathymetry data in a 10-by-13 km area around the Isles of Shoals. Ground truth data were collected by lowering and raising cameras from the seabed; observations were divided into bare substrate, kelp habitat, red turf macroalgae habitat, and intermediate "mixed" macroalgae habitat classes, and used to train a Random Forest model. The model classified benthic habitats with 71% accuracy. Depth, median summer sea surface temperature, vector ruggedness measure, and slope were among the most important variables in classifying kelp habitat. This approach improves upon previous modelling and monitoring methods by expanding the depth range and total amount of area that can be assessed, while also addressing the importance of temperature in mediating substrate competition between kelps and other macroalgae. It may be generalizable to the Gulf of Maine and to other regions where kelp habitats face similar stressors and may aid in identifying healthy habitats for conservation.

Anthropogenic pressures have altered food webs by accelerating the addition of novel species through range expansions and species introductions. Globally, the Gulf of Maine (GoM) has been one of the most rapidly changing bodies of water, leading to shifts in the abundances of two foundational species-the blue mussel Mytilus edulis, and the slipper limpet, Crepidula fornicata. M. edulis populations have declined due to rapid warming and anthropogenic induced threats, while C. fornicata populations have increased. Here, we examine the relationship of M. edulis and C. fornicata as prey species to a suite of three native (the cancrid crabs Cancer borealis and Cancer irroratus, and the lobster Homarus americanus) and two non-native crustacean predators in the GoM (the green crab Carcinus maenas and the Asian shore crab Hemigrapsus sanguineus) through a series of preference studies, nutritional assays and handling experiments. While all predators except H. sanguineus had better nutritional condition when feeding on M. edulis than when feeding on C. fornicata, only C. maenas and H. americanus exhibited a strong preference for M. edulis. Handling times for M. edulis were significantly longer than those for C. fornicata, suggesting that some predators are willing to expend more energy for nutritious prey. Populations of crustaceans are under threat from warming water temperatures, ocean acidification, and introduced species, all of which can increase their energetic demands. While C. fornicata is a poor diet replacement for M. edulis, it is ubiquitous and abundant in benthic environments and may serve as a useful supplemental prey in times of M. edulis and other prey scarcity.

The decline of important reef building corals has motivated the development of habitat suitability models used to identify optimal locations for coral restoration. In the Florida Keys habitat suitability models incorporate coarse spatial data sampled over large areas, resulting in recommended outplant sites at distant locations, making it logistically difficult and expensive to access and regularly monitor. Restoration efforts to date show that outplanting success can vary widely within a limited space, necessitating improved predictive abilities of coral outplant success at high spatial resolutions within a restoration site. With the advent of Structure-from-Motion image reconstruction, fine-scale, site specific, digital terrain models can be created to support habitat suitability model development. In this study, generalized linear mixed models used extracted seafloor terrain attributes and environmental variables to identify within site locations of high Acropora cervicornis growth and healthy coral cover of long-term outplants. Percent healthy coral cover significantly decreased after two years of outplantation. The submodel of corals exclusively less than two years old was unable to identify environmental conditions associated with higher healthy cover. For all corals, outplant recommendations for higher healthy cover are in deeper waters, away from the coast, in less rough terrain, and closer to the reef edge. Model results for growth support these recommended outplant sites, in addition to concave locations near high slope relief. Finally, our results also indicate that marine heat waves, but especially marine cold waves negatively correspond with coral growth, and high wind events positively correspond with coral growth. These model results provide a basis for further endeavors in modeling endangered organismal success, which are vulnerable to minute differences in local environmental conditions. ### Competing Interest Statement The authors have declared no competing interest.

AbstractThe impacts of global change—from shifts in climate to overfishing to land use change—can depend heavily on local abiotic context. Building an understanding of how to downscale global change scenarios to local impacts is often difficult, however, and requires historical data across large gradients of variability. Such data are often not available—particularly in peer reviewed or gray literature. However, these data can sometimes be gleaned from casual records of natural history—field notebooks, data sheet marginalia, course notes, and more. Here, we provide an example of one such approach for the Gulf of Maine, as we seek to understand how environmental context can influence local outcomes of region‐wide shifts in subtidal community structure. We explore a decade of hand‐drawn algal cover maps around Appledore Island made by Dr. Art Borror while teaching at the Shoals Marine Lab. Appledore's steep wave exposure gradient—from exposed to the open ocean to fully protected—provides a living laboratory to test interactions between global change and local conditions. We then recreate Borror's methods two and a half decades later. We show that overfishing‐driven urchin outbreaks in the 1980s were slowed or stopped by wave exposure and benthic topography. Similarly, local variation appears to have curtailed current invasions by filamentous red algae. Last, some formerly dominant kelps have disappeared over the past 40 years—an observation verified by subtidal surveys. Global change is altering life in the seas around us. While underutilized, solid natural history observations stand as a key resource for us to begin to understand how global change will translate to the heterogeneous mosaic of life in a future Gulf of Maine and other ecosystems around the world.

Ridge Flank Hydrothermal Systems have discrete pockets of fluid discharge that mimic climate-induced ocean warming. Unlike traditional hydrothermal fluids, those discharged by Ridge Flank Hydrothermal Systems have a chemical composition indistinguishable from background water, enabling evaluation of the effect of warming temperature. Here we link temperature and terrain variables to community composition and biodiversity by combining remotely operated vehicle images of vent and non-vent zone communities with associated environmental variables. We show overall differences in composition, family richness, and biodiversity between zones, though richness and diversity were only significantly greater in vent zones at one location. Temperature was a contributing factor to observed greater biodiversity near vent zones. Overall, our results suggest that warming in the deep sea will affect species composition and diversity. However, due to the diverse outcomes projected for ocean warming, additional research is necessary to forecast the impacts of ocean warming on deep-sea ecosystems. A study of opportunity on deep-sea Ridge Flank Hydrothermal Systems, which host low-temperature, chemically insignificant venting, suggests that warmer temperatures are a contributing factor to changing community composition and higher biodiversity.

Artificial environments have hard surfaces positioned at different orientations that attract a wide diversity of sessile invertebrate species, forming fouling communities. Fouling communities play a large role in the spread of a species introduction, as organisms gain purchase in artificial environments and use them as a stepping-stone into neighboring natural systems. These man-made systems harbor a large diversity of species, and there are often vastly different communities present on both vertical and horizontal surfaces. We used a species abundance dataset at three time periods spanning 30 years, collected from fouling panels in a high tidal flow estuary at the mouth of the Piscataqua River (New Castle, New Hampshire) in the southern Gulf of Maine to measure differences in community composition between horizontal and vertical panels. Early successional communities on settlement panels were photographed one year after installment in late summer to capture the highest diversity seasons. Organisms were then identified to the lowest taxonomic level, and communities on the underside of horizontal and vertical panels were compared. Vertical and horizontal communities from 1980 and 2004 were similar, while those from 2009 statistically differed from one another. Vertical and horizontal surfaces in the 1980s were dominated by the blue mussel Mytilus edulis, which declined in later years, and was replaced by invasive ascidians. Species diversity increased between the 1980 and 2004 sampling, but declined by the 2009 sampling. This study captures changes of species composition in early successional communities on vertical and horizontal surfaces at a single locale from three time periods spanning over 30 years.

AbstractInvasive species can disrupt food webs by altering the abundance of prey species or integrating into the food web themselves. In the Gulf of Maine, there have been a suite of invasions that have altered the composition of the benthic ecosystem. These novel prey species can potentially benefit native predators depending on their nutritional value and relative abundance. We measured feeding instances of the native blood star, Henricia sanguinolenta, and changes in the seasonal abundances of invasive ascidian prey species. Results indicate that H. sanguinolenta forages optimally, as the blood star will prey on invasive ascidians when in high abundance, but feed on other species during periods of scarcity. Further, our study shows that blood stars prey on a wider variety of species than was previously known, such as small bivalves and barnacles. Additionally, we compared growth and reproduction of sea stars fed different combinations of invasive ascidians (Diplosoma listerianum or Botrylloides violaceus) or a native sponge (Haliclona oculata). Sea stars grew more on the native diet when compared with the invasive ascidian species, and D. listerianum appeared to be a superior quality food source when compared with B. violaceus. By comparing our data with historical data, we determined that there was a dramatic increase in sea star populations between 1980 and 2011, but then populations decreased by almost half from 2011 to 2016–2017. These data suggest that while invasive ascidians may have helped sea star populations at one point, sea stars are declining without their native food source.

Disease outbreaks and mass mortality events in terrestrial and marine taxa are increasing as a result of human influence, pollution, and climate change. Sea Star Wasting has become more common on the west coast of the United States, with increasing numbers of asteroids affected since 2013. Rising temperatures have traditionally been linked to mass mortality events, but other factors such as the effect of diet on individual susceptibility to Sea Star Wasting have not been examined. Here, we use laboratory experiments to test the frequencies of SSW sign onset and death of a native sea star when exposed to diets of two invasive ascidians. We show that prey identity affects mortality rates of affected predators. Sea stars that consumed Diplosoma listerianum experienced SSW signs, but 0% mortality, whereas animals that consumed Botrylloides violaceus, or no prey experienced 41% mortality. We expect that invader induced shifts in community species composition will increase the frequency of these mass die offs on both the west and east coasts.

Deep sea canyons and seamounts are topographically complex features that are considered to be biological hotspots. Anthropogenic pressures related to climate change and human activities are placing the species that inhabit these features at risk. Though studies have examined species composition on seamounts and canyons, few have compared communities between them, and even fewer studies have examined how species’ abundances correlate with environmental conditions or geomorphology. Consequently, this study compares species composition, community structure, and environmental variables between Northwest Atlantic continental margin canyons and seamounts along the New England Seamount Chain. Geoforms were also related to the occurrence of phyla and biodiversity. Overall, there was a significant difference in species composition between canyons and seamounts with sponges, corals, sea urchins and seastars contributing heavily to observed differences. Environmental conditions of temperature and salinity and the seafloor property slope contributed significantly to communities observed on seamounts, while substrate, depth and salinity contributed significantly to canyon communities. Abundances were significantly higher in canyons, but taxonomic richness, evenness, and diversity were all greater on seamounts. In an era where climate change and human activity have the potential to alter environmental parameters in the deep sea, it is important to examine factors that influence the spatial distribution of deep-sea benthic communities.

Aim Environmental variables are strongly tied to species occurrence and population growth, but approaches to predicting the location of deep-sea species or their ability to withstand a changing environment stem primarily from presence data. We coupled environmental data with observed densities of deep-sea habitat-forming corals and sponges to determine the environmental variables and geomorphology that contributed best to their occurrence. Location Northwest Atlantic. Time period 2013 and 2014. Major taxa studied Deep-sea coral and sponge communities. Methods Multivariate and univariate analyses were used to determine significant environmental contributors to densities of genera and families of corals and sponges. We then assessed the relationship of densities of genera and families of corals and sponges with environmental variables found to be significant contributors to their occurrence and to geomorphology. Results Sponge and coral genera and families were influenced by different environment variables. Temperature, salinity and dissolved oxygen contributed to the occurrence of sponges, whereas seafloor properties of slope and substrate contributed to the occurrence of corals. Although individuals of corals and sponges were observed across a range of a contributing environmental variable, high densities were observed only in very narrow ranges. Main conclusions Geomorphic setting is an effective approach for discerning the associations of coral with seabed features. High densities of coral and sponge genera and families restricted to narrow environmental ranges might be at greater risk of local extinction. Differences in the occurrence of coral and sponge genera and families with environmental conditions suggest that they will differentially respond to predicted environmental changes. As conditions in the deep sea change with ongoing changes in climate, population expansion might be limited owing to suboptimal conditions, and established populations might persist but might have fewer individuals or species, which might lead to a loss in biodiversity.

We extracted and analyzed microplastics (MP) in archived sediment cores from Great Bay Estuary (GBE) in the Gulf of Maine region of North America. Results indicated that MP are distributed in GBE sediments, 0-30 cm, at an average occurrence of 116 ± 21 particles g-1 and that morphology varies by site and depth. Analysis by sediment depth and age class indicated that MP accumulation increased over several decades but recently (5-10 years) has likely begun to decrease. Hydrodynamic and particle transport modeling indicated that bed characteristics are a more controlling factor in MP distribution than typical MP properties and that the highest accumulation likely occurs in regions with weaker hydrodynamic flows and lower bed shear stress, e.g., eelgrass meadows and along fringes of the Bay. These results provide a baseline and predictive understanding of the occurrence, morphology, and sedimentation of MP in the estuary.

Benthic quadrat surveys using 2-D images are one of the most common methods of quantifying the composition of coral reef communities, but they and other methods fail to assess changes in species composition as a 3-dimensional system, arguably one of the most important attributes in foundational systems. Structure-from-motion (SfM) algorithms that utilize images collected from various viewpoints to form an accurate 3-D model have become more common among ecologists in recent years. However, there exist few efficient methods that can classify portions of the 3-D model to specific ecological functional groups. This lack of granularity makes it more difficult to identify the class category responsible for changes in the structure of coral reef communities. We present a novel method that can efficiently provide semantic labels of functional groups to 3-D reconstructed models created from commonly used SfM software (i.e., Agisoft Metashape) using fully convolutional networks (FCNs). Unlike other methods, ours involves creating dense labels for each of the images used in the 3-D reconstruction and then reusing the projection matrices created during the SfM process to project semantic labels onto either the point cloud or mesh to create fully classified versions. When quantitatively validating the classification results we found that this method is capable of accurately projecting semantic labels from image-space to model-space with scores as high as 91% pixel accuracy. Furthermore, because each image only needs to be provided with a single set of dense labels this method scales linearly making it useful for large areas or high resolution-models. Although SfM has become widely adopted by ecologists, deep learning presents a steep learning curve for many. To ensure repeatability and ease-of-use, we provide a comprehensive workflow with detailed instructions and open-sourced the programming code to assist others in replicating our methodology. Our method will allow researchers to assess precise changes in 3-D community composition of reef habitats in an entirely novel way, providing more insight into changes in ecological paradigms, such as those that occur during coral-algae shifts.

Eastern oyster Crassostrea virginica populations have been declining steadily over the past several decades across the North American East coast. The Great Bay Estuary (GBE), located in New Hampshire, is experiencing this loss and restoration efforts have been put into effect. This paper characterizes larval abundances of settled spat and two early stages of C. virginica, D-hinge and veliger, in GBE from 2018 to 2020. Abundances are compared based on date of sampling, year, collection site, and the physicochemical data recorded on each sampling date. It was found that overall, D-hinge larval abundances have declined significantly from 2018 to 2020, whereas veliger abundances have remained steady or increased. Although the physicochemical factors are known to play a role in larval abundance, very little significance was found, suggesting future study may need to be modified to include a broader range of factors (e.g., more temporal sampling). This study indicates that both D-hinge, veliger, and spat settlement occur in GBE before sampling traditionally has started (June), suggesting an earlier than previously thought first spawn of C. virginica in GBE. This finding can be used to enhance restoration efforts as it suggests that spat brought in to augment current sites of active restoration should be released earlier in the season and that recruitment devices should be deployed before the previously thought first spawn of each season.

This case study applied the Coastal and Marine Ecological Classification Standard (CMECS) to initial characterization of a deep-sea seamount by combining observations from a remotely operated vehicle (ROV) and information derived from multibeam sonar bathymetry and backscatter. Spatial segmentation of the multibeam bathymetry was done using algorithms based on defining bathymorphons resulting in six classes: flats, slopes, ridges, valleys, shoulders, and footslopes. These classes were modified to delineate CMECS "Level 1" geoform units for Gosnold Seamount. Further segmentation of landforms was completed using textural analysis of the sonar backscatter mosaic of the seamount to identify segments of the same landform type with similar reflectivity texture. All of the ROV dive video of the seafloor was analyzed manually to create a spreadsheet of 933 georeferenced annotations of organisms and associated substrate types. The dominant sediment type over each 50 m segment of the ROV track was also classified using substrate unit terminology from CMECS into four classes: bedrock (10% of ROV track), fine unconsolidated sediments on bedrock (84%), coral rubble (1%), and sand (5%). Eleven genera of corals, two classes of sponges, and four classes of echinoderms were observed along the track, with glass sponges dominating the annotation and abundance counts. Nominal regression revealed that depth, temperature, and sediment type were significant predictors of individual coral along the ROV track (P<.001, P<.001, P<.001, respectively). In contrast, slope, sediment type and dissolved oxygen were significant predictors of sponge distribution along the track. In summary the application of CMECS to Gosnold Seamount provided a useful systematic framework for structuring geoform, substrate, and biotic classification of benthic habitat. Using this standard, in combination with the semiautomated seafloor segmentation approach utilized, can provide a consistent and reproducible habitat classification approach for large regions and facilitate comparison of habitats among features.

Benthic quadrat studies requiring time-intensive manual image annotation are currently a critical component of assessing the health of coral reefs. Patch-based image classification using convolutional neural networks (CNNs) can automate this task by providing sparse labels, but remain computationally inefficient. This work extends the idea of automatic image annotation by using fully convolutional networks (FCNs) to provide dense labels through semantic segmentation. We present an improved version the Multilevel Superpixel Segmentation (MSS) algorithm, which repurposes existing sparse labels for images by converting them into the dense labels necessary for training a FCN automatically. Our implementation-Fast-MSS-is demonstrated to perform considerably faster than the original without sacrificing accuracy. To showcase the applicability to benthic ecologists, we validate this method using the Moorea Labeled Coral (MLC) dataset as a benchmark. FCNs are evaluated by comparing their predictions on test images with the corresponding ground-truth sparse labels. Our results indicate that FCNs' perform with accuracies that are suitable for many ecological applications, and can increase even further when trained on dense labels augmented with additional sparse labels provided by a patch-based image classifier. The contributions of this study help move the field of benthic ecology towards more efficient monitoring of coral reefs through entirely automated processes.

Foundation species like macroalgae provide habitat for large numbers of animals. The spatial structure between branches or thalli can act as a refuge from larger predators and can affect the number and distribution of inhabitant species. Most metrics for habitat architecture are based on 2-dimensional measurements, but habitats are 3-dimensional. We report a new method, spherical space analysis, for characterizing the 3-dimensional volume distribution by size of interstitial spaces for 3 species of macroalgae (seaweed) with distinct architectures. This analysis gives the distribution of volumes within a foundation species that are inaccessible by an idealized spherical organism-an 'inaccessible volume curve'. A second product is an 'inaccessible surface area curve'. We incorporated abundances and size ranges of meso-invertebrates into spherical space analysis to predict predator-prey interactions as a function of the relationship between inaccessible volume and area and the size of predators and prey. The results show that filamentous forms of macroalgae have more smaller interstitial volume and area than branched or blade forms of macroalgae that support a larger number of smaller meso-invertebrates. The model suggests that the spatial structure of macroalgae affects predator-prey interactions with a greater number of smaller spaces providing more refuge. This was particularly apparent for kelp. Spherical space analysis provides a mechanism for understanding how the spatial architecture of a macroalgal environment mediates the network of feeding interactions occurring within it. This can have implications for restoration efforts, as the morphology and density of foundation species are integral in the maintenance of communities.