Deep learning for coastal resource conservation: automating detection of shellfish reefs

Ridge, Justin T.; Gray, Patrick C.; Windle, Anna E.; Johnston, David W.

doi:10.1002/rse2.134

Cited by 26 publications

(21 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Many authors have called for comprehensive coastal ecosystem models that deal with component complexities, especially biodiversity [72]. Higher resolution studies combined with deep learning techniques are beginning to address the complexities of reef communities [78,79].…”

Section: Toxicity: Shoreline and Underwatermentioning

confidence: 99%

Coastal Remote Sensing: Merging Physical, Chemical, and Biological Data as Tailings Drift onto Buffalo Reef, Lake Superior

et al. 2021

View full text Add to dashboard Cite

On the Keweenaw Peninsula of Lake Superior, two stamp mills (Mohawk and Wolverine) discharged 22.7 million metric tonnes (MMT) of tailings (1901–1932) into the coastal zone off the town of Gay. Migrating along the shoreline, ca. 10 MMT of the tailings dammed stream and river outlets, encroached upon wetlands, and contaminated recreational beaches. A nearly equal amount of tailings moved across bay benthic environments into critical commercial fish spawning and rearing grounds. In the middle of the bay, Buffalo Reef is important for commercial and recreational lake trout and lake whitefish production (ca. 32% of the commercial catch in Keweenaw Bay, 22% along southern Lake Superior). Aerial photographs (1938–2016) and five LiDAR and multispectral over-flights (2008–2016) emphasize: (1) the enormous amounts of tailings moving along the beach; and (2) the bathymetric complexities of an equal amount migrating underwater across the shelf. However, remote sensing studies encounter numerous specific challenges in coastal environments. Here, we utilize a combination of elevation data (LiDAR digital elevation/bathymetry models) and in situ studies to generate a series of physical, chemical, and biological geospatial maps. The maps are designed to help assess the impacts of historical mining on Buffalo Reef. Underwater, sand mixtures have complicated multispectral bottom reflectance substrate classifications. An alternative approach, in situ simple particle classification, keying off distinct sand end members: (1) allows calculation of tailings (stamp sand) percentages; (2) aids indirect and direct assays of copper concentrations; and (3) permits determinations of density effects on benthic macro-invertebrates. The geospatial mapping shows how tailings are moving onto Buffalo Reef, the copper concentrations associated with the tailings, and how both strongly influence the density of benthic communities, providing an excellent example for the International Maritime Organization on how mining may influence coastal reefs. We demonstrate that when large amounts of mine tailings are discharged into coastal environments, temporal and spatial impacts are progressive, and strongly influence resident organisms. Next steps are to utilize a combination of hi-resolution LiDAR and sonar surveys, a fish-monitoring array, and neural network analysis to characterize the geometry of cobble fields where fish are successful or unsuccessful at producing young.

show abstract

Section: Toxicity: Shoreline and Underwatermentioning

confidence: 99%

Coastal Remote Sensing: Merging Physical, Chemical, and Biological Data as Tailings Drift onto Buffalo Reef, Lake Superior

et al. 2021

View full text Add to dashboard Cite

show abstract

“…They provide ecosystem services, including essential nursery habitat, food and shelter for fish and marine organism, carbon sequestration, sea bottom stabilization, improved water quality, and shoreline protection. However, these landscapes are at risk from the combined stress of climatic disasters, such as typhoons and rainfalls, and direct human-driven changes, such as the release of pollutants, making effective management and conservation increasingly crucial [67]. Capsule ANNs were proposed in [84] to address several limitations that CNNs present, such as the invariance caused by pooling and the inability to understand spatial relationships between features.…”

Section: B Ecosystem Preservationmentioning

confidence: 99%

“…Fast R-CNN improves on the R-CNN by only performing CNN forward computation on the image as a whole [87]. Other examples focusing on automatic species detection based on DL classification of underwater images are [67] where the authors detect and delineate oyster reefs, and [68], [69] where annotation of marine coral species is performed.…”

Section: B Ecosystem Preservationmentioning

confidence: 99%

Sustainable Marine Ecosystems: Deep Learning for Water Quality Assessment and Forecasting

et al. 2021

View full text Add to dashboard Cite

An appropriate management of the available resources within oceans and coastal regions is vital to guarantee their sustainable development and preservation, where water quality is a key element. Leveraging on a combination of cross-disciplinary technologies including Remote Sensing (RS), Internet of Things (IoT), Big Data, cloud computing, and Artificial Intelligence (AI) is essential to attain this aim. In this paper, we review methodologies and technologies for water quality assessment that contribute to a sustainable management of marine environments. Specifically, we focus on Deep Leaning (DL) strategies for water quality estimation and forecasting. The analyzed literature is classified depending on the type of task, scenario and architecture. Moreover, several applications including coastal management and aquaculture are surveyed. Finally, we discuss open issues still to be addressed and potential research lines where transfer learning, knowledge fusion, reinforcement learning, edge computing and decision-making policies are expected to be the main involved agents.

show abstract

“…Object-based image analysis (OBIA) has been used to effectively classify UAS-based maps of shallow water environments such as seagrass meadows and hard bottom habitats (Chabot et al, 2018;Ventura et al, 2018). Many of these algorithms are now established tools within geospatial software packages, but some habitat classifications may require more complex applications of machine learning such as neural networks to train a computer how to identify target habitats or metrics (Casado et al, 2015;Ridge et al, 2020). In this context, UAS can be applied to the development of training data for machine learning systems, as well as for conducting rapid validation sampling that reduces human impacts on restored systems (Gray et al, 2018).…”

Section: Data Types and Usesmentioning

confidence: 99%

Unoccupied Aircraft Systems (UAS) for Marine Ecosystem Restoration

Ridge

Johnston

2020

Front. Mar. Sci.

Self Cite

View full text Add to dashboard Cite

Assessing, implementing and monitoring ecosystem restoration can be a labor intensive process, often short term (<3 years), and potentially destructive to the habitat. Advances in remote sensing technology are generating rapid, non-destructive methods for siting, executing and monitoring restoration efforts, particularly in fragile marine environments. Unoccupied aircraft systems (UAS), or drones, are a highly flexible method for accessing and remote sensing ecosystems with on-demand capabilities, greater resolution than sensors from satellites and occupied aircraft, and the ability to cover large areas quickly. With the variety of platforms and payloads available, UASs are providing a suite of tools for conservation practitioners to properly plan marine ecosystem restoration projects and evaluate their success. Both conventional and specialized sensors coupled with image processing techniques can be used to gauge impact to and recovery of entire ecological communities. For example, high-resolution, multispectral imaging allows for discernment of population changes across trophic levels, concurrent with the discrimination of species (including rare) across a landscape, and detection of vegetation stress. Structure from Motion photogrammetric processing provides centimeter-scale three-dimensional models of habitat structure to measure ecologically significant aspects like rugosity and assess their change through time. Water quality around a broad impacted area can be remotely monitored via a number of payloads before and after restoration. Additionally, specially designed payloads can be used to manually disperse seeds or materials for restoration applications without disturbing the habitat. UASs have increasing potential to reduce the costs (both time and money) associated with restoration efforts, making site assessment and long-term, broad-scale monitoring more achievable. Here we present a review of the applications of UASs in marine ecosystem restoration with an overview of the special considerations of using this technology in the marine environment.

show abstract

Deep learning for coastal resource conservation: automating detection of shellfish reefs

Cited by 26 publications

References 31 publications

Coastal Remote Sensing: Merging Physical, Chemical, and Biological Data as Tailings Drift onto Buffalo Reef, Lake Superior

Coastal Remote Sensing: Merging Physical, Chemical, and Biological Data as Tailings Drift onto Buffalo Reef, Lake Superior

Sustainable Marine Ecosystems: Deep Learning for Water Quality Assessment and Forecasting

Unoccupied Aircraft Systems (UAS) for Marine Ecosystem Restoration

Contact Info

Product

Resources

About