An AI Breakthrough in Ocean Monitoring: Satellite Imagery Provides an Unprecedented View of Global Vessel Activity

Global Fishing Watch releases optical imagery mapping, tripling its vessel detections and enhancing transparency in human activity at sea.

A clearer view of human activity across the ocean is now available through global satellite imagery. Our latest dataset features vessel detections based on 10-meter resolution optical images from the European Space Agency’s Sentinel-2 satellites, which capture high-quality images of the ocean.

Satellite map showing vessel activity along the coasts of China, Korea, and Japan, with colored lines indicating high to low ship density; green-yellow areas represent more vessels, blue fewer, per the color scale.
Before - Sentinel-1
A satellite heat map showing vessel detections around eastern China, the Korean Peninsula, and Japan; yellow indicates higher vessel concentrations, while green/blue indicates fewer, with a legend at the bottom right.
After - Sentinel-2

The number of vessel detections and area covered has increased significantly through use of optical imagery from the Sentinel-2 satellites, as can be seen here along the coast of East Asia.

Our new approach allows us to detect about three times more vessels than we identified in our landmark Nature paper, and provide 30 percent more coverage of the ocean,  extending visibility to all coastal waters where the majority of human activity takes place. The detailed optical imagery combined with our novel AI approach allows the detection of smaller vessels–less than 10 meters long in some cases–while also estimating the length, orientation and speed for nearly every moving vessel at sea.

Our workflow employs AI end-to-end for vessel detection and classification. These AI models, known as artificial neural networks, consist of interconnected nodes or “neurons” in a layered structure through which the information flows in order to recognize patterns.

The detection model was trained on over 11,000 manually reviewed vessel examples across thousands of Sentinel-2 images captured around the world. The model scans image tiles to detect vessels, looking for characteristic features that it has learned. These features might consist of geometric shapes, pixel brightness and patterns in the water from vessel movement.

A satellite image shows multiple vessels at sea, with a zoomed-in circle highlighting three adjacent vessels. Text on the image reads: Activity: Three adjacent vessels. Detection length: 151 meters.
Satellite image of a fishing vessel 37 meters long, moving at 4 knots, engaged in bottom trawling. The boat is leaving a faint trail in deep blue water, shown in a zoomed-in circular inset.
Satellite image showing islands and surrounding water with several vessels. A zoomed-in circle highlights two adjacent ships with text: “Activity: Two adjacent vessels. Detection length: 190 meters.”.
Aerial view of a sand dredging vessel in the ocean, leaving a visible trail. Text on the image notes the vessels activity, length (194 meters), and speed (3 knots), with magnified circles highlighting the ship.
Satellite image showing an icebreaking vessel, 239 meters long, moving at 0.3 knots through icy waters near a coastline. A magnified circle highlights the ship’s position along a broken path in the ice.
Aerial image showing tracks of two vessels pair trawling in the ocean, with annotated details: left vessel 41 meters, right vessel 36 meters, speed 3 knots. A magnified circle highlights the vessels’ parallel paths.

Previously, we mapped vessel activity using Sentinel-1 detections based on synthetic aperture radar (SAR) imagery at 20-meter resolution, which is excellent at detecting metal objects but had challenges identifying vessels made of wood and fiberglass. Our new mapping uses Sentinel-2 optical imagery at 10-meter resolution, allowing for the detection of smaller vessels made from all types of materials.

Black irregular shadow on a textured, gray surface, with the shadow resembling the profile of a creatures head or an abstract shape. The background appears grainy and rough.
20m Radar
A satellite view of a coastal area with towns, green vegetation, a bay, and turquoise shoreline; boats and white wakes are visible scattered across the dark blue sea.
10m Optical

A regional view of Saint-Tropez, France, showing the difference between Sentinel-1 radar and Sentinel-2 optical imagery.

The high-quality optical images also show features on the water such as wake patterns, allowing for the calculation of vessel speed and orientation. Here, you can see the difference between a fast-moving vessel, with a clear pattern behind it in the water, and one moving at a slower speed with a more subtle wake pattern.

The detailed information contained in the optical imagery can be used to map not only the presence of vessels…

Mapa mundial que muestra las zonas oceánicas por nivel de seguimiento de buques: las zonas azules son de seguimiento público, las zonas amarillas no son de seguimiento público. Las regiones costeras de Asia, África y Sudamérica tienen más zonas sin seguimiento.

but also their activity, as illustrated below:

A color scale legend shows vessel speed in knots from 1 (pale yellow) to 10 (blue), with markers at 1, 3, 6, and 10 knots. Vessel length scale ranges from 5m to 500m on the right.

A view of vessel activity along the coast of the Yellow Sea, showing the estimated speed, length and orientation of detected vessels.

A closer view of activity within the Port of Tianjin.

A view of vessel activity through the Strait of Malacca, showing the estimated speed, length and orientation of detected vessels.

A closer view of activity within the Strait of Malacca.

We are working on next generation models to identify different types of vessel activity across the ocean, including bottom trawling, sand dredging and transshipment, opening a new era of mapping possibilities.

We also rely on AI to identify and remove objects that are not vessels, such as offshore infrastructure, and rocks or icebergs that look like vessels. Detections are limited, however, by light and weather—since optical satellites rely on sunlight and clear skies, detections are only possible during the day and when the area is not obscured by clouds or haze.

The ability to monitor vessel activity from space in key locations like marine protected areas, exclusive economic zones and areas overseen by regional fisheries management organizations, will allow for advances in research and management.

And while the 10-meter resolution is still too coarse to fully map small-scale fishing, Sentinel-2 detections have been used to support multiple analyses related to regional small-scale fisheries and demonstrated their potential as a valuable addition to the limited vessel tracking data.

With this novel technology, we are adding a new powerful feature to the Global Fishing Watch map. Now, each vessel detection based on optical imagery includes a visual thumbnail, representing an area of 1 square kilometer, showing the detected vessel. The thumbnails are especially helpful for visually confirming a vessel is present or understanding its context.

See for yourself!

Explore the map now
Scroll to Top