New Crowdin updates (#2698)

New translations

libs/i18n-labels/en/datasets.json

@@ -995,28 +995,28 @@
     }
   },
   "public-fixed-infrastructure-filtered-v1.1": {
-    "name": "Fixed infrastructure",
-    "description": "SAR identified fixed infrastructure",
+    "name": "Offshore Fixed Infrastructure (SAR, Optical)",
+    "description": "<h2>Overview</h2> <p>Offshore fixed infrastructure is a global dataset that uses AI and machine learning to detect and classify structures throughout the world’s oceans.</p> <p>Classification labels (oil, wind, and unknown) are provided, as well as confidence levels (high, medium, or low) reflecting our certainty in the assigned label. Detections can be filtered and colored on the map using both label and confidence level.<em></em>The data is updated on a monthly basis, and new classified detections are added at the beginning of every month. Viewing change using the timebar is simple, and allows anyone to recognize the rapid industrialization of the world’s oceans. For example, you can easily observe the expansion of wind farms in the North and East China Seas, or changes in oil infrastructure in the Gulf of Mexico or Persian Gulf.</p> <p>By overlaying the existing map layers, you can explore how vessels interact with oil and wind structures, visualise the density of synthetic aperture radar (SAR) and Visible Infrared Imaging Radiometer Suite (VIIRS) vessel detections around infrastructure, or determine which marine protected areas (MPAs) contain wind, oil, or other infrastructure types. These are only examples of the types of questions we can now ask. Offshore fixed infrastructure is a first of its kind dataset that not only brings to light the extensive industrialization of our oceans, but enables users across industries to use this information in research, monitoring and management.</p> <h2>Use cases</h2> <ul> <li>Maritime domain awareness</li> <ul> <li>Infrastructure locations can support maritime domain awareness, and understanding of other activities occurring at sea.</li> <li>Infrastructure data supports assessments of ocean industrialization, facilitating monitoring of areas experiencing build-up or new development</li> </ul> <li>Monitoring vessels</li> <ul> <li>Infrastructure locations can be used to analyse the behaviour of vessels associated with infrastructure, including grouping vessels based on their interaction with oil and wind structures.</li> <li>Interactions between vessels and infrastructure can help quantify the resources required to support offshore industrial activity</li> <li>The impacts of infrastructure on fishing, including attracting or deterring fishing, can be analysed.</li> </ul> <li>Marine protected areas (MPAs) and marine spatial planning</li> <ul> <li>During the planning stage in the designation of new protected areas, knowing the location of existing infrastructure will be vital to understand which stakeholders shall be included in the consultation process, to understand potential conflicts, and identify easy wins.</li> </ul> <li>Environmental impacts</li> <ul> <li>Infrastructure locations can be used to help detect marine pollution events, and to differentiate between types of pollution events (e.g. pollution from vessels versus pollution from platforms)</li> </ul> </ul> <h2>Caveats</h2> <ul> <li><strong>Sentinel-1 and Sentinel-2 satellites do not sample most of the open ocean.</strong></li> <ul> <li>Most industrial activity happens relatively close to shore.</li> <li>The extent and frequency of SAR acquisitions is determined by the mission priorities.</li> <li>For more info see: https://www.nature.com/articles/s41586-023-06825-8/figures/5</li> </ul> <li><strong>We do not provide detections of infrastructure within 1 km of shore</strong></li> <ul> <li>We do not classify objects within 1 km of shore because it is difficult to map where the shoreline begins, and ambiguous coastlines and rocks cause false positives.</li> <li>The bulk of industrial activities, including offshore development with medium-to-large oil rigs and wind farms, occur several kilometers from shore.</li> </ul> <li> <strong>False positives can be produced from noise artifacts.</strong> </li> <ul> <li>Rocks, small islands, sea ice, radar ambiguities (radar echoes), and image artifacts can cause false positives</li> <li>Detections in some areas including Southern Chile, the Arctic, and the Norwegian Sea have been filtered to remove noise.</li> </ul> <li><strong>Spatial coverage varies over time, which can produce different detections results year on year - <a target=\"_blank\" href=\"https://share.cleanshot.com/yG0qfF\"> <span style=\"color:rgb(0, 0, 0);\">Example</span> </a></strong> </li> <ul> <li>Infrastructure detentions from 2017-01-01 to near real time are available, and updated on a monthly basis.</li> </ul> <li> <strong>Labels can change over time</strong> </li> <ul> <li>The label assigned to a structure is the greatest predicted label averaged across time. As we get more data, the label may change, and more accurately predict the true infrastructure type.</li> </ul> <li><strong>Global datasets aren’t perfect</strong></li> <ul> <li>We’ve done our best to create the most accurate product possible, but there will be infrastructure that isn’t detected, or has been classified incorrectly. This will be most evident when working at the project level.</li> <li>We strongly encourage users to provide feedback to the research team so that we may improve future versions of the model. All feedback is greatly appreciated.</li> </ul> </ul> <h2>Methods</h2> <h3>SAR imagery</h3> <p>We use SAR imagery from the Copernicus Sentinel-1 mission of the European Space Agency (ESA) [1]. The images are sourced from two satellites (S1A and S1B up until December 2021 when S1B stopped operating, and S1A only from 2022 onward) that orbit 180 degrees out of phase with each other in a polar, sun-synchronous orbit. Each satellite has a repeat-cycle of 12 days, so that together they provide a global mapping of coastal waters around the world approximately every six days for the period that both were operating. The number of images per location, however, varies greatly depending on mission priorities, latitude, and degree of overlap between adjacent satellite passes. Spatial coverage also varies over time [2]. Our data consist of dual-polarization images (VH and VV) from the Interferometric Wide (IW) swath mode, with a resolution of about 20 m.</p> <p>[1] <a target=\"_blank\" href=\"https://sedas.satapps.org/wp-content/uploads/2015/07/Sentinel-1_User_Handbook.pdf\"> <span style=\"color:rgb(0, 0, 0);\">https://sedas.satapps.org/wp-content/uploads/2015/07/Sentinel-1_User_Handbook.pdf</span> </a> </p> <p>[2]<a target=\"_blank\" href=\"https://sentinels.copernicus.eu/web/sentinel/missions/sentinel-1/observation-scenario\"> <span style=\"color:rgb(0, 0, 0);\"></span> <span style=\"color:rgb(0, 0, 0);\">https://sentinels.copernicus.eu/web/sentinel/missions/sentinel-1/observation-scenario</span> </a> </p> <h3>Infrastructure detection by SAR</h3> <p>Detecting infrastructure with SAR is based on the widely used Constant False Alarm Rate (CFAR) algorithm, an anomaly detection method conceived for detecting ships in synthetic aperture radar images, that has been modified to remove non-stationary objects. This algorithm is designed to search for pixel values that are unusually bright (the targets) compared to those in the surrounding area (the sea clutter). This method sets a threshold based on the pixel values of the local background (within a window), scanning the whole image pixel-by-pixel. Pixel values above the threshold constitute an anomaly and are likely to be samples from a target.</p> <h3>Infrastructure classification</h3> <p>To classify every detected offshore infrastructure, we used deep learning and designed a ConvNet based on the ConvNeXt architecture. A novel aspect of our deep learning classification approach is the combination of SAR imagery from Sentinel-1 with optical imagery from Sentinel-2. From six-month composites of dual-band SAR (VH and VV) and four-band optical (RGB and NIR) images, we extracted small tiles for every detected fixed infrastructure, with the respective objects at the center of the tile. A single model output includes the probabilities for the specified classes: wind, oil, unknown, lake maracaibo, and noise.</p> <h3>Filtering</h3> <p>GFW post-processed the classified SAR detections to reduce noise (false positives), remove vessels, exclude areas with sea ice at high latitudes, and incorporate expert feedback. We used a clustering approach to identify detections across time (within a 50 m radius) that were likely the same structure but their coordinates differed slightly, and assigned them the greatest average predicted label of the cluster. We also filled in gaps for fixed structures that were missing in one timestep but detected in the previous and following timesteps, and dropped detections appearing in a single timestep. Finally, the dataset underwent extensive manual review and editing by researchers and industry experts in order to refine the final product, and provide the most accurate dataset possible.</p> <h3>Data field descriptions</h3> <p>Each detection has a unique individual identifier (<em>detection_id</em>). A six-month image composite is used in the classification, therefore the <em>detection_date</em> represents the middle of the six month period. This helps to remove non-stationary objects (i.e. vessels), and avoid confusion in the model if a structure is being built, or there isn’t adequate imagery available. <em>structure_id</em> allows you to track a structure through time. There are therefore many <em>detection_id</em> (one for each month the structure is detected) for each <em>structure_id</em>. Labels of <em>wind</em> and <em>oil </em>represent any wind or oil related structure respectively. <em>Unknown</em> represents a structure that is not oil or wind related, such as bridges or navigational buoys. </p> <p>Label confidence levels of ‘High’. ‘Medium’ and ‘Low’ are assigned to each structure, and are conditional on where the detections fell in relation to the boundaries of manually developed wind and oil polygons, and whether the label has changed from the previous month. The <em>label_confidence</em> field can be used to filter analysis. </p> <h2>Resources, code and other notes</h2> <p>Two repos are used in the automation process, both of which are private, and should not be shared publicly.</p> <p>Detection and classification: https://github.com/GlobalFishingWatch/sentinel-1-ee/tree/master</p> <p>Clustering and reclassification: https://github.com/GlobalFishingWatch/infrastructure-post-processing</p> <p>All code developed for the paper, Paolo, F.S., Kroodsma, D., Raynor, J. et al. Satellite mapping reveals extensive industrial activity at sea. Nature 625, 85–91 (2024). https://doi.org/10.1038/s41586-023-06825-8, including SAR detection, deep learning models, and analyses is open source and freely available at https://github.com/GlobalFishingWatch/paper-industrial-activity.</p> <h2>Sources data and citations</h2> <p>Copernicus Sentinel data 2017-current</p> <p>Lujala, Päivi; Jan Ketil Rød &amp; Nadia Thieme, 2007. 'Fighting over Oil: Introducing A New Dataset', Conflict Management and Peace Science 24(3), 239-256</p> <p>Sabbatino, M., Romeo, L., Baker, V., Bauer, J., Barkhurst, A., Bean, A., DiGiulio, J., Jones, K., Jones, T.J., Justman, D., Miller III, R., Rose, K., and Tong., A., Global Oil &amp; Gas Infrastructure Features Database Geocube Collection, 2019-03-25, https://edx.netl.doe.gov/dataset/global-oil-gas-infrastructure-features-database-geocube-collection, DOI: 10.18141/1502839</p> <h2>License</h2> <p>Non-Commercial Use Only. The Site and the Services are provided for Non-Commercial use only in accordance with the CC BY-NC 4.0 license. If you would like to use the Site and/or the Services for commercial purposes, please contact us.</p> <h2>Global Fishing Watch metadata</h2> <p>Infrastructure development methods should reference the paper:</p> <p>Paolo, F.S., Kroodsma, D., Raynor, J. et al. Satellite mapping reveals extensive industrial activity at sea. Nature 625, 85–91 (2024). https://doi.org/10.1038/s41586-023-06825-8</p> <p>All code developed for the paper, including SAR detection, deep learning models, and analyses is open source and freely available at https://github.com/GlobalFishingWatch/paper-industrial-activity. All the data generated and used by these scripts can reference the following data repos:</p> <p>Analysis and Figures: https://doi.org/10.6084/m9.figshare.24309475</p> <p>Training and Evaluation: https://doi.org/10.6084/m9.figshare.24309469</p>",
     "schema": {
       "label": {
-        "keyword": "label",
+        "keyword": "Label",
         "enum": {
-          "oil": "oil",
-          "wind": "wind",
-          "unknown": "unknown"
+          "oil": "Oil",
+          "wind": "Wind",
+          "unknown": "Unknown"
         }
       },
-      "structure_id": "structure_id",
+      "structure_id": "Structure ID",
       "label_confidence": {
-        "keyword": "label_confidence",
+        "keyword": "Confidence",
         "enum": {
-          "high": "high",
-          "low": "low",
-          "medium": "medium"
+          "high": "High",
+          "low": "Low",
+          "medium": "Medium"
         }
       },
-      "structure_end_date": "structure_end_date",
-      "structure_start_date": "structure_start_date"
+      "structure_end_date": "End date",
+      "structure_start_date": "Start date"
     }
   },
   "public-global-all-tracks": {