Datasets and Code: Apparent Fishing Effort

Global Fishing Watch’s flagship dataset is apparent fishing effort based on transmissions broadcast using the automatic identification system (AIS). In 2018, we published the first global assessment of commercial fishing activity in the journal Science. Our research found that fishing is both widespread—occurring throughout approximately 50 percent of the ocean—and highly concentrated, with more than half of fishing activity occurring in just 0.5 percent of the ocean. 

Fishing is also minimally affected by seasons but strongly affected by culture, with the largest drops occurring during weekends, holidays and the annual Chinese summer moratorium. 

Today we receive over 110 million AIS messages each day and continually update our technology and algorithms to improve our ability to monitor global commercial fishing. This entire dataset, dating back to 2012, can be explored on our map, through our public APIs and R package, or downloaded from our data download portal.

How it works

Using cloud computing, machine learning and public vessel registry information, we analyze tens of millions of AIS positions each day to map global apparent fishing effort. Producing such a map involves two key steps: 

    •  Identification of fishing vessels in the AIS data
    •  Detection of fishing activity

Processing AIS data

AIS is a GPS-like device that large ships use to broadcast their position in order to avoid collisions. Each year, hundreds of thousands of AIS devices broadcast vessel location along with other information on vessel identity, course and speed. Ground stations and satellites pick up this information, meaning a ship’s movements can be followed even in the most remote parts of the ocean.

Every day we receive raw AIS data from our AIS providers. These data are first run through a series of algorithms designed to filter out corrupt or incomplete records and assign additional information to each AIS message, such as the distance from shore, depth and time since the vessel’s previous AIS position. At this point, the AIS data is ready to be used by our machine learning models.

Detecting apparent fishing activity

We use two convolutional neural networks (CNN)—a form of machine learning model—to help us classify fishing vessels and predict when they are fishing. We refer to these models as our “vessel characterization” and “fishing detection” models, respectively. The details of both CNN models are described in detail in the supplementary materials of our 2018 Science paper, “Tracking the global footprint of fisheries.”

Before we can map fishing activity, we must first identify fishing vessels in the AIS data. We accomplish this by combining our comprehensive database of vessel registry information with the output of our vessel characterization model—predicting features like geartype and size—and use the best available information for each vessel. Next, we estimate where and when each vessel is fishing based on its movement patterns. 

We manually labeled over a thousand vessel tracks to train our fishing detection model to learn what fishing movements look like. This model predicts a score for every AIS position in our database to distinguish fishing positions from non-fishing positions. When our fishing detection model scores an AIS position as a fishing position, the time associated with that AIS position is considered apparent fishing activity.

We manually labeled over a thousand AIS vessel tracks, whether or not a vessel was fishing at individual AIS positions. We then used this training data to teach our fishing detection model to identify other fishing positions in the entire AIS dataset.

Summarizing apparent fishing effort

After identifying fishing vessels and detecting fishing positions in the AIS data, apparent fishing effort can be calculated for any area by summarizing the fishing hours for all fishing vessels in that area. To generate maps of apparent fishing effort we then “rasterize” the AIS positions by placing them into a grid and calculating the total fishing activity in each grid cell.

A dark blue world map showing apparent fishing hours detected with AIS for the time period 2012 through 2024, with a gradient scale indicating apparent fishing hours per square kilometer in the ocean from low (blue) through green and yellow to high (white)
A global map shows apparent fishing hours per square kilometer from the current release of our fishing effort dataset, covering 2012-2024. © 2025 Global Fishing Watch

Using vessel identity information, such as vessel class and flag State, allows us to describe fishing activity by specific groups of vessels. Because fishing activity is assigned to individual AIS points, we can make these rasters of AIS-based apparent fishing activity at any spatial and temporal resolution. 

Caveats and limitations

While AIS provides a revolutionary way to monitor global commercial fishing activity, there are several important limitations and caveats. First, AIS data includes only a small fraction—approximately 100,000—of the world’s estimated 3.3 million fishing vessels. Coverage is much higher for larger vessels, with less than 1 percent of vessels under 12 meters represented, approximately 20 percent for vessels between 12 to 24 meters, and up to 90 percent for vessels larger than 24 meters. The International Maritime Organization mandates AIS for most vessels larger than 36 meters, and vessels broadcasting AIS are predominantly from upper and upper-middle income countries.   

Another key caveat is that not every AIS message that is broadcast is recorded. A receiver must be in range to record a message, so satellites must be overhead or dynamic AIS receivers (receivers carried onboard vessels) must be nearby. Terrestrial receivers also exist, but only receive signals near shore. Additionally, AIS messages can interfere with each other in areas of high vessel density and AIS devices vary in broadcast strength and frequency. As a result, fewer AIS positions are received by vessels operating in certain parts of the world, limiting the effectiveness of our machine learning models and our ability to detect apparent fishing effort.

A pair of world maps show the quality of two classes of reception of satellite AIS data in 2023, with a gradient scale indicating the number of satellite positions per day.
A global view of the reception quality of satellite AIS data, represented as the average number of AIS positions you can expect to receive from a single AIS device per day in different parts of the world.

Finally, as more vessels have adopted AIS and more receivers have started recording AIS messages, the amount of activity in the AIS dataset has increased. The rise in activity in the early years of the AIS dataset (2012-2016) at least partially reflects the increase in satellite and terrestrial receivers and should not simply be interpreted as an increase in fishing activity, though that may have happened in some areas. Furthermore, around 2022, we began to receive data through dynamic AIS receivers, which are receivers carried onboard vessels. This new data source substantially improved our reception of AIS messages in some areas of high vessel density, particularly in the waters around China and Southeast Asia. Increases in apparent fishing effort should be interpreted in light of these changes. We are working toward the release of a dataset of global reception quality to allow researchers and others to account for these changes.

Access the data

Anyone with an internet connection can trace the movements of over 100,000 commercial fishing boats, along with their name and flag State. All users of our map can access our near real-time data to create heat maps to visualize patterns of commercial fishing activity, view tracks of individual vessels and overlay information like the locations of marine protected areas or countries’ exclusive economic zones. The same data is also available in our public APIs and R package.

Our AIS-based static apparent fishing effort data can be downloaded via our data download portal. Data are available in the following formats: 

    • Daily apparent fishing hours by flag state and gear type at 100th degree resolution
    • Monthly apparent fishing hours by flag state and gear type at 10th degree resolution
    • Daily apparent fishing hours by MMSI at 10th degree resolution

These static datasets are also available on Google’s BigQuery platform from the following publicly available dataset: global-fishing-watch.fishing_effort_v3 

Scroll to Top