MMDEC: Multimodal Maritime Dataset on the English Channel

Description The MMDEC dataset contains data from various sources and sensors (AIS, satellite images, meteorology, oceanography, ports locations, marine protected areas,...) within an Area of Interest (AOI) covering the western Celtic Sea, the English Channel, and a part of the North Sea, and a 3-month time from July 1, 2023, to October 1, 2023 DOI identifier https://doi.org/10.5281/zenodo.17491518 Contributors Tristan AVERTY (Ph.D), IRENav, École navale, Brest, France Ioannis NASIOS, NodalPoint Systems, Athens, Greece Cyril RAY (Ph.D), IRENav, École navale, Brest, France Nikos PILIOURAS (Ph.D), NodalPoint Systems, Athens, Greece File format ZIP / CSV / Parquet / JPG / GeoJSON / SHP / HTML / PDF Number of files 21 files Total size ~ 11.5 GB Last update 2025-10-31 Funding Dataset built within the MUSIT project. This project has received funding from the European Union’s Horizon Europe Framework Programme (HORIZON) under Grant Agreement No. 101182585 🧩 List of files 🌍 AOI.geojson (Area of interest) Satellite-related data and AI predictions 🎞 Dataset_S1_chunks.zip (Satellite products chunks) 🛰 Dataset_S1_chunks_corners.zip (Coordinates of satellite products chunks) 🤖 Dataset_S1_SatShipAI_outputs.zip (AI detection & classification) 🚫 Dataset_S1_SatShipAI_excluding_areas.zip (Excluding areas) AIS-related data 📍 Dataset_AIS_POS.parquet (AIS position messages) 📋 Dataset_AIS_SPEC.parquet (AIS specifications messages) 🚢 Dataset_AIS_ShipTypes.csv (AIS ship types) 🤿 Dataset_BATHYMETRY.parquet (Bathymetry) ⚡ Dataset_CABLES.parquet (Telecommunication and power submarine cables) 💧 Dataset_CMEMS_PHY.parquet (Oceanography / physical) 🌊 Dataset_CMEMS_WAV.parquet (Oceanography / wave-based) 🌦 Dataset_ERA5.parquet (Meteorology) 🐠 Dataset_MPA.parquet (Marine protection areas) ⚓ Dataset_PORTS.parquet (Ports location) 🛟 Dataset_SECMAR.parquet (French maritime operations) 🛣 Dataset_TSS.parquet (Traffic Separation Schemes) 🏝 Dataset_SEAS.parquet (Sea areas) 🌪 Dataset_WIND_FARMS.parquet (Wind farms location) 🗺 VISUALIZATION.html (Interactive map) 🌍 AOI.geojson (Area of interest) File name AOI.geojson File size ~ 3 KB SRID WGS84 (ESPG:4326) Volume 1 polygon Description Polygon that is the AOI considered for all the following data Licence © CC BY 4.0 import geopandas as gpd AOI = gpd.read_file("AOI.geojson") Satellite-related data and AI predictions For the construction of the satellite dataset, data from the Copernicus Sentinel-1 Synthetic Aperture Radar (SAR) satellites were used. Specifically, the Ground Range Detected (GRD) products were employed, using the VV and VH polarization channels acquired in the Interferometric Wide (IW) swath mode. These products provide dual-polarized SAR measurements with high spatial resolution, suitable for maritime monitoring and ship detection applications. The satellite dataset is composed of the 4 archives describes as follows. 🎞 Dataset_S1_chunks.zip (Satellite products chunks) File name Dataset_S1_chunks.zip File size ~ 10.5 GB Source Copernicus Sentinel-1 data. https://browser.dataspace.copernicus.eu/ Spatial coverage AOI.geojson Temporal coverage From 2023-07-01 to 2023-10-01 Volume 230 subfolders Licence © CC BY-NC-SA 4.0 This folder is structured by satellite names as subfolders, each containing JPEG chunk images with dimensions of 800×800×3 (note that some edge chunks have different shapes). To create these chunks, the full satellite product is first processed using ESA’s SNAP software and then scaled to the 0–255 range with data type int8. The three channels are constructed as (VV, VH, (VV+VH)/2). Only chunks located over the sea and within the Area of Interest (AOI) are retained. Using JPEG image chunks reduces the data volume from approximately 2 GB per product to about 50 MB, while producing images in a format suitable for neural network processing. Dataset_S1_chunks/ ------------------ - S1A_IW_GRDH_1SDV_20230701T062352_20230701T062417_049226_05EB4B_F66A/ - i0j2.jpg - i0j3.jpg - ... - S1A_IW_GRDH_1SDV_20230701T062417_20230701T062442_049226_05EB4B_2A99/ - i0j0.jpg - i0j11.jpg - ... - ... 🛰 Dataset_S1_chunks_corners.zip (Coordinates of satellite products chunks) File name Dataset_S1_chunks_corners.zip File size ~ 5 MB SRID WGS84 (EPSG:4326) Volume 230 CSV files Licence © CC BY-NC-SA 4.0 This folder contains 1 CSV file per satellite product, where each file contains the above chunk images corner geocoordinates. The columns of the CSV files are described as follows: Feature Description chunk_name The name of the image (".jpg" excluded) lat1 Top Left corner Latitude of the image chunk in decimal degrees Lon1 Top Left corner Longitude of the image chunk in decimal degrees lat2 Top Right corner Latitude of the image chunk in decimal degrees Lon2 Top Right corner Longitude of the image chunk in decimal degrees lat3 Bottom Right corner Latitude of the image chunk in decimal degrees Lon3 Bottom Right corner Longitude of the image chunk in decimal degrees lat4 Bottom Left corner Latitude of the image chunk in decimal degrees Lon4 Bottom Left corner Longitude of the image chunk in decimal degrees Dataset_S1_chunks_corners/ -------------------------- - S1A_IW_GRDH_1SDV_20230701T062352_20230701T062417_049226_05EB4B_F66A.csv - S1A_IW_GRDH_1SDV_20230701T062417_20230701T062442_049226_05EB4B_2A99.csv - ... 🤖 Dataset_S1_SatShipAI_outputs.zip (AI detection & classification) File name Dataset_S1_SatShipAI_outputs.zip File size ~ 746 KB SRID WGS84 (EPSG:4326) Volume 231 CSV files Licence © CC BY-NC-SA 4.0 This folder contains 1 CSV file per satellite product, where each file includes the AI model’s estimations, specifically the geolocation of detected ships, detection confidence, estimated ship length, and predicted ship type along with its associated probability. All predictions were generated using the trained models deployed on the SatShipAI platform. Further details regarding the satellite product processing, employed models, performance evaluation, and related aspects can be found at https://satshipai.eu/ and https://www.mdpi.com/2079-9292/14/18/3648. A detailed description of these features is provided in the table below. Feature Description Example lon Ship Longitude in decimal degrees -6.24421996958622 lat Ship Latitude in decimal degrees 48.1442381705964 X Ship location in the X axis on the Sentinel-1 product 23236 Y Ship location in the Y axis on the Sentinel-1 product 9673 detection_prob The probability output from the detection model 0.884398 ship_length The estimated ship length in meters 291 ship_class The estimated ship type, one of: [Cargo, Fishing, Other, Passenger, Tanker] Tanker ship_class_prob The probability output from the classification model 0.6179871 Furthermore, the additional CSV file named extra_info.csv contains 1 record for every satellite product with the following features: Feature Description product_name The name of the Sentinel-1 satellite SAR product product_width The Width of the product (number of pixels) product_height The Height of the product (number of pixels) begindatetime The date and time of the product IncAng_long_multi 1 Multiplier for any longitude value with the product IncAng_lat_multi 1 Multiplier for any latitude value with the product intercept 1 A value to be added North_azimuth The North azimuth vertical_pixelDistance The Vertical Pixel Distance lat11 Top Left corner Latitude in decimal degrees Lon11 Top Left corner Longitide in decimal degrees lat12 Top Right corner Latitude in decimal degrees Lon12 Top Right corner Longitide in decimal degrees lat22 Bottom Right corner Latitude in decimal degrees Lon22 Bottom Right corner Longitide in decimal degrees lat21 Bottom Left corner Latitude in decimal degrees Lon21 Bottom Left corner Longitide in decimal degrees 1To be used for estimating the incidence angle wherever within the satellite product Dataset_S1_SatShipAI_outputs/ ----------------------------- - S1A_IW_GRDH_1SDV_20230701T062352_20230701T062417_049226_05EB4B_F66A.csv - S1A_IW_GRDH_1SDV_20230701T062417_20230701T062442_049226_05EB4B_2A99.csv - ... - extra_info.csv 🚫 Dataset_S1_SatShipAI_excluding_areas.zip (Excluding areas) File name Dataset_S1_SatShipAI_excluding_areas.zip File size ~ 6 KB SRID WGS84 (EPSG:4326) Volume 5 files Licence © CC BY 4.0 For more reliable model estimations by SatShipAI, at the post-processing step a set of exclusions was applied: Exclude estimations on land Exclude estimations at distance lower than 500 m to land Exclude estimations outside our AOI. Estimations within wind farm areas were excluded from the analysis. For the dataset AOI, three wind farm regions were removed. These areas are provided in the Dataset_S1_SatShipAI_excluding_areas.zip archive, which contains both a SHP and a GeoJSON file, each representing the same AOI for convenience. AIS-related data 📍 Dataset_AIS_POS.parquet (AIS position messages) File name Dataset_AIS_POS.parquet File size ~ 451 MB Source French Naval Academy SRID WGS84 (EPSG:4326) Spatial coverage AOI.geojson Temporal coverage From 2023-07-01 to 2023-10-01 Volume 19 014 229 messages for 25 130 unique MMSI Description Parquet file containing position messages (AIS of type 1, 2, 3, 18, 19 and 27) - https://www.navcen.uscg.gov/ais-messages Licence © CC BY-NC-SA 4.0 Attribute Data type Description Date datetime Date when the report was generated by the electronic position system Source object Type of AIS receiver "eee-land" : Terrestrial "eee-sat" : Satellite MessageType int32 1 : Scheduled position report; Class A shipborne mobile equipment 2 : Assigned scheduled position report; Class A shipborne mobile equipment 3 : Special position report, response to interrogation; Class A shipborne mobile equipment 18 : Standard position report for Class B shipborne mobile equipment to be used instead of Messages 1, 2, 3 19 : Extended position report for Class B shipborne mobile equipment; contains additional static information 27 : Class A and Class B "SO" shipborne mobile equipment outside base station coverage Mmsi uint32 MMSI number of the ship sending the position report NavigationStatus float64 0 : under way using engine 1 : at anchor 2 : not under command 3 : restricted maneuverability 4 : constrained by her draught 5 : moored 6 : aground 7 : engaged in fishing 8 : under way sailing 9 : reserved for future amendment of navigational status 10 : reserved for future amendment of navigational status 11 : power-driven vessel towing astern (regional use) 12 : power-driven vessel pushing ahead or towing alongside (regional use) 13 : reserved for future use 14 : AIS-SART (active), MOB-AIS, EPIRB-AIS 15 : undefined (default) Latitude float64 Latitude (in degrees) Longitude float64 Longitude (in degrees) PositionAccuracy bool Position accuracy (PA) in accordance with the rules: True (PA <= 10 m) False (PA > 10 m or default) CourseOverGroundDegrees float32 Course over ground (in degrees) 511 indicates not available (default) SpeedOverGround float32 Speed over ground (in knots) RateOfTurn float64 0 to +126° : turning right at up to 708 deg per min or higher 0 to -126° : turning left at up to 708 deg per min or higher + 127° : turning right at more than 5 deg per 30 s (No TI available) -127 : turning left at more than 5 deg per 30 s (No TI available) -128 (80 hex) indicates no turn information available (default). TrueHeadingDegrees float64 True heading (in degrees) 511 indicates not available (default) chunk_folder object Name of the folder (=name of the satellite product) in Dataset_S1_chunks.zip that contains chunks in which the ship declaring its position can be found. NaN if such a satellite product does not exist. id_chunk object List of chunk names (without the ".jpg") in which the ship declaring its position can be found. NaN if such chunks do not exist. import pyarrow.parquet as pq POS = pq.read_table("Dataset_AIS_POS.parquet").to_pandas() 📋 Dataset_AIS_SPEC.parquet (AIS specifications messages) File name Dataset_AIS_SPEC.parquet File size ~ 210 MB Source French Naval Academy SRID WGS84 (ESPG:4326) Spatial coverage AOI.geojson Temporal coverage From 2023-07-01 to 2023-10-01 Volume 13 558 007 messages for 23 958 unique MMSI Description Parquet file containing static and voyage related messages (AIS of type 5 and 24) - https://www.navcen.uscg.gov/ais-messages Licence © CC BY-NC-SA 4.0 Attribute Data type Description Date datetime Date when the report was generated by the electronic position system Source object Type of AIS receiver "eee-land" : Terrestrial "eee-sat" : Satellite MessageType int32 5 : Scheduled static and voyage related vessel data report, Class A shipborne mobile equipment 24 : Additional data assigned to an MMSI Part A: Name Part B: Static Data Mmsi uint32 MMSI number of the ship sending the position report ImoNumber float64 0 : not available (default) 0000000001-0000999999 : not used 0001000000-0009999999 : valid IMO number; 0010000000-1073741823 : official flag state number. CallSign object 6 or 7 characters @@@@@@@ : not available (default) Craft associated with a parent vessel should use “A” followed by the last 6 digits of the MMSI of the parent vessel. VesselName object Maximum 20 characters @@@@@@@@@@@@@@@@@@@@ : not available (default) The name should be as shown on the station radio license. ShipType float64 Types described in the Dataset_AIS_ShipType.csv DimensionToBow float64 In meters DimensionToStern float64 In meters DimensionToPort float64 In meters DimensionToStarboard float64 In meters Draught10thMetres float64 Draught in 1/10 m 255 : draught of 25.5 m or greater 0 : not available (default) Destination object Maximum 20 characters @@@@@@@@@@@@@@@@@@@@ : not available EtaMonth float64 Estimated month of arrival (1-12) 0 : not available (default) EtaDay float64 Estimated day of arrival (1-31) 0 : not available (default) EtaHour float64 Estimated hour of arrival (0-23) 24 : not available (default) EtaMinute float64 Estimated minute of arrival (0-59) 60 : not available (default) PositionFixType float64 Type of electronic position fixing device 0 : undefined (default) 1 : GPS 2 : GLONASS 3 : combined GPS/GLONASS 4 : Loran-C 5 : Chayka 6 : integrated navigation system 7 : surveyed 8 : Galileo 9-14 : not used 15 : internal GNSS import pyarrow.parquet as pq SPEC = pq.read_table("Dataset_AIS_SPEC.parquet").to_pandas() 🚢 Dataset_AIS_ShipTypes.csv (AIS ship types) File name Dataset_AIS_ShipTypes.csv File size ~ 3 KB Source https://coast.noaa.gov/data/marinecadastre/ais/VesselTypeCodes2018.pdf Volume 256 lines Description List of the 256 different ship types in AIS messages Licence © CC BY 4.0 Attribute Data type Description Code datetime Code that can be found in the ShipType column in Dataset_AIS_SPEC.parquet Type object Corresponding ship main type (Other, Fishing, Tug Tow, Pleasure Craft/Sailing, Passenger, Cargo, Tanker or not available) Additional object Additional description import pandas as pd SHIP_TYPE = pd.read_csv("Dataset_AIS_ShipTypes.csv", sep=";") ⚓ Dataset_PORTS.parquet (Ports location) File name Dataset_PORTS.parquet File size ~ 35 KB Source Claude Merrien (2021). Worldwide list of seaports. https://doi.org/10.12770/59ab5f6f-79ea-425d-830e-be5ecdb7bdbe SRID WGS84 (ESPG:4326) Spatial coverage AOI.geojson Volume 399 ports Description Location of ports within the AOI Licence © CC BY-NC-SA 4.0 Attribute Data type Description Country object Country to which the port is affiliated LOCODE object UNECE or ERS 5-character coding Name object Name of the port (or the city where the port is) Latitude float64 Latitude (in degrees) Longitude float64 Longitude (in degrees) Status object Status of the port indicating whether it is referenced by Unece (UNECE), by the Electronic Recording and Reporting System (ERS) or by both (UNECE/ERS) National_C object For French ports, national codification Group object For French ports, indicate whether it it a group of ports ('Y') or not ('N') geometry geometry Geometry in order to import it directly with GeoPandas import geopandas as gpd PORTS = gpd.read_parquet("Dataset_PORTS.parquet") 🛟 Dataset_SECMAR.parquet (French maritime operations) File name Dataset_SECMAR.parquet File size ~ 154 KB Source Ministère de la Transition écologique (2025). Opérations coordonnées par les CROSS. https://www.data.gouv.fr/datasets/operations-coordonnees-par-les-cross/ SRID WGS84 (ESPG:4326) Spatial coverage AOI.geojson Temporal coverage From 2023-07-01 to 2023-10-01 Volume 1769 operations Description Operations of the Regional Operating Surveillance and Rescue Centres (CROSS) within the AOI and the 3-month time period Licence © CC BY 4.0 All the variables are described here (in French) : https://mtes-mct.github.io/secmar-documentation/schema.html#operations As usual, we have added a geometry column in order to import it directly with GeoPandas. import geopandas as gpd SECMAR = gpd.read_parquet("Dataset_SECMAR.parquet") 🛣 Dataset_TSS.parquet (Traffic Separation Schemes) File name Dataset_TSS.parquet File size ~ 14 KB Source SHOM (2021). Dispositifs de séparation du trafic. https://www.data.gouv.fr/datasets/dispositifs-de-separation-du-trafic-1 SRID WGS84 (ESPG:4326) Spatial coverage AOI.geojson Volume 19 polygons Description This dataset represents the areas separating two traffic lanes or a traffic lane from a coastal navigation area or the central area of a roundabout that intersect with the AOI Licence © CC BY 4.0 Attribute Data type Description inspireId object INSPIRE identification code of the TSS nom object TSS designation reference object Name of the reference text referenceL object Link to the reference text geometry geometry Geometry in order to import it directly with GeoPandas import geopandas as gpd TSS = gpd.read_parquet("Dataset_TSS.parquet") 🏝 Dataset_SEAS.parquet (Sea areas) File name Dataset_SEAS.parquet File size ~ 4 MB Source Flanders Marine Institute (2018). IHO Sea Areas, version 3. Available online at https://www.marineregions.org/ https://doi.org/10.14284/323 SRID WGS84 (ESPG:4326) Spatial coverage AOI.geojson Volume 3 polygons Description This dataset represents the boundaries of the major seas that intersect the AOI. The source for the boundaries is the publication 'Limits of Oceans & Seas, Special Publication No. 23' published by the IHO in 1953. Licence © CC BY-NC-SA 4.0 Attribute Data type Description NAME object Sea area designation geometry geometry Geometry in order to import it directly with GeoPandas import geopandas as gpd SEAS = gpd.read_parquet("Dataset_SEAS.parquet") 🌪 Dataset_WIND_FARMS.parquet (Wind farms location) File name Dataset_WIND_FARMS.parquet File size ~ 15 KB Source EMODnet Human Activities, Energy, Wind Farms (2025). https://emodnet.ec.europa.eu/geonetwork/srv/api/records/8201070b-4b0b-4d54-8910-abcea5dce57f?language=all SRID WGS84 (ESPG:4326) Spatial coverage AOI.geojson Volume 3 polygons Description This dataset represents the wind farms that were producing or being constructed within the time period Licence © CC-BY 4.0 Attribute Data type Description COUNTRY object Country to which the wind farm is affiliated NAME object Wind farm designation N_TURBINES int64 Number of windmills in the wind farm POWER_MW float64 Power (in MW) of the wind farm STATUS object Status of the wind farm START_YEAR int64 Year in which the wind farm entered its current status COAST_DIST float64 Distance (in meters) between the coastline and the wind farm AREA_SQKM float64 Area (in square kilometers) of the wind farm geometry geometry Geometry in order to import it directly with GeoPandas import geopandas as gpd WF = gpd.read_parquet("Dataset_WIND_FARMS.parquet") 🗺 VISUALIZATION.html (Interactive map) File name VISUALIZATION.html File size ~ 116 MB Description Interactive map made with Folium that represents a subset of all the data collected within this dataset between 2023-07-08 06:10 and 2023-0708 06:20 Licence © CC-BY 4.0