| Popis: |
The goal of this deliverable is to assess availability of FAIR spatial data about human activity using global datasets, and to identify obstacles (knowledge gaps) to mapping their corresponding environmental pressures at the scale of the Atlantic Ocean basin to support ecosystem-based management. It also provides advice towards improving mapping based on machine learning/artificial intelligence (AI). WP1 & 7 identified the sectors affecting the marine environment, the pressures they create, and the ecological characteristics affected. Eighteen sectors, with 19 associated pressures that could impact 25 different ecosystem components were identified. In this deliverable, only the widely distributed sectors relevant across the whole ocean are considered: fishing, shipping, aquaculture, oil exploitation, telecommunications, scientific exploration and seabed minerals. Fishing intensity by main groups of fishing gears has been estimated by supervised classification of Automatic Identification System (AIS) data. These approaches have been able to distinguish between fishing and routing activity of individual vessels, while assigning up to 7 main fishing gear types. These are aggregated here into pelagic and bottom fishing activities to match the grouping of ecosystem components in these two groups. AIS data is most useful in the high seas since in coastal areas, smaller vessels do not require AIS, except in some regions of Europe. At the Canary Current Sea CS pelagic fishing shows the highest intensities around the Canaries and Cabo Verde archipelagos and all the way from western Sahara to Guinea-Bissau. On the contrary, bottom trawling concentrates closer to the coast off the coasts of northern Morocco and from western Sahara to Guinea-Bissau. At North Mid Atlantic Ridge CS scale, AIS data shows pelagic fishing is widely distributed except in the northern part, and bottom fishing seems to be concentrated on the Islands. At South Mid Atlantic Ridge CS, scale AIS data shows a wide distribution of pelagic fishing whereas bottom fishing seems to be poorly covered. At Brazilian Shelf CS scale, bottom and pelagic fishing are widely distributed on the area. Shipping activities can be also inferred from AIS data. AIS systems reporting position through a satellite system are mandatory in all passenger ships irrespective of size, all vessels > 300 MT (gross) transiting international routes, and any cargo ship > 500 MT transiting within national waters. Global patterns reveal that highest shipping intensity is concentrated in coastal areas with narrow routes connecting Asia, Africa and Europe. Connectivity between these continents and America is broader with a medium intensity over large areas. This can be observed in the Europe connection with North America, Europe connection to South America, South Africa connection with South America and in the connection of Asia with North America and with Australia. At the Atlantic scale, the areas of highest shipping intensity are concentrated in European coastal waters. These higher intensity paths cross the southern Celtic Sea, Canary Current, and western Benguela Current case studies. The rest of the case studies also contain relatively high intensity paths, but not as extreme. The large area covered by high intensity paths in Canary Current, Benguela Current and South Brazilian Shelf case studies is noteworthy. The North Mid Atlantic has moderate intensity paths covering most of its area with two higher intensity paths crossing it. The Norwegian Sea and the South Mid Atlantic Ridge case studies include large areas of low shipping intensity. We focus on three main sources of pollution: 1) Marine litter (widespread diffuse input), and in particular plastics (microplastics and macroplastics) and 2) Aquaculture platforms, as proxy of marine pollution sources (point source). feed, leading to major environmental concerns. 3) River discharges and nutrients (in between widespread and point source). Although river discharge and associated nutrients are a natural pressure in the coastal environment, anthropogenically polluted rivers can significantly alter the health of marine ecosystems, especially as metallic pollutants are associated with sediments. Open litter data is sparse and focused mainly on European waters. Continuous Plankton Recorder, which can also be used to sample plastics, covers not only European waters, but also an important part of Northwest Atlantic area. After revising multiples sources of river discharges and nutrients discharges the WaterGAP 2.2d model and the Global River Water Quality Archive have been identified as the best sources of data at the Atlantic Ocean and global scales. Aquaculture data is very limited even in European open datasets and it can be addressed applying machine learning to satellite data. Other seabed related data include oil exploitation, telecommunications, scientific exploration and seabed minerals. These data are integrated for European waters in EMODnet Seabed Habitats. In the Norwegian Sea, Celtic Sea, Benguela Current and South Brazilian Shelf additional local sources of oil exploitation data were available. In terms of telecommunications data, published data are predominantly landing stations and schematic routes. Regarding accurate spatial data on scientific surveys, only a few openly available sources have been identified at national and regional levels, and metadata are not standardised to be able to easily identify specific locations. Seabed mineral potential has been collated by EMODnet Seabed Habitats for most of the case study areas, except for the Benguela Current where local data were provided. Additional seabed mineral data were also provided for the South Brazilian Shelf, North Mid Atlantic Ridge and Benguela Current case study areas. The term big data was coined to capture the meaning of the emerging trend in large and heterogeneous data exploitation. In addition to its sheer volume, big data exhibit other unique characteristics as compared with traditional data. For instance, big data are commonly unstructured and require real-time analysis. This development calls for new computer science related system architectures for data acquisition, transmission, storage, and large-scale data processing mechanisms. Big data techniques enhanced by machine learning methods can increase the use of such data and their applicability to ecosystem-based management. Machine learning has already proven its potential in marine science. It has been applied, for example, to fisheries forecasting, automatic classification of plankton samples, identification of schools by species of fish in acoustic survey data, litter classification and litter forecasting. However, if AI methods are not fit for purpose, then trade-offs to accomplish appropriate performances can be missed or overfitting can lead to over confidence on AI capacity if proper validation and ground truth verification is not performed. This report highlights big data and machine learning-based datasets that are available to perform assessments at the level of the whole Atlantic Ocean. These datasets focus on the activities that impact open-ocean marine environments (shipping, fishing and litter) and coastal areas (river discharges and flows, and the growing aquaculture activity). The differentiation between fishing and non-fishing (e.g. travel) activities of fishing vessels as well as distinguishing the fishing gear is needed to go from mapping activities to pressures. For example, a fishing vessel en route to a fishing site will generate different levels and types of noise and emissions than it will whilst fishing. Noise and emissions will also differ depending on the fishing gear being used. Furthermore, the highest direct pressure exerted on the ecosystem occurs during fishing due to the removal of fish and bycatch, or litter generation from losing fishing gear and/or onboard activities. Members of ICES working group WGSHIP are actively working on the development of a conceptual framework for mapping the pressures and impacts of shipping. |
