AIS and the Challenges of Tracking Vessels at Sea

At Global Fishing Watch, we hear it all the time: “Tracking commercial fishing vessels from satellites is such a great idea, and it seems so easy!” In fact, we’ve received a few questions from our readers asking us why this isn’t just a simple hack of publicly available data.

On one hand, it sounds simple. Large commercial fishing vessels already transmit their location and identity via the Universal Shipborne Automatic Identification System (AIS), so all we need to do is plot the signals on a chart or map of the ocean to identify their tracks, and compare those to patterns corresponding to fishing activity. Voilà! A global snapshot of all the vessels over a certain size that are fishing on the ocean.

Well, unfortunately there’s more to the equation. Much more. In one random sample, we observed more than 127,000 vessels broadcasting over a 24-hour period. With that many vessels using AIS on the ocean, and signals refreshing as often as every five seconds, there are billions of data points and terabytes of data to feed into our computer systems. Making sense of them poses many challenges.

The root of the problem is that tracking technology was designed to be used between ships at sea——a safety feature that lets vessels “shout out” their presence to avoid collisions. Using those messages for tracking their behavior poses three basic types of challenges.

1] AIS signals are noisy and not designed for this purpose.

2] Vessel identities can be cryptic

3] Simply plotting a vessel’s track doesn’t tell us what that vessel is doing.

Here are some specific examples of these challenges with links to posts that describe how our analysts are overcoming them:

Noise:

AIS systems transmit radio signals in the Maritime VHF band, and the airwaves are inherently noisy. Think of all the static, or white noise, coming through on a poorly tuned radio station. The first step to understanding what’s being transmitted is to remove all the signals that don’t make sense, all the white noise. That in itself is tricky because you have to figure out exactly how much noise to remove in order to see the patterns of real ships without removing so many ambiguous signals that you miss real signals from real vessels.

AIS Noise comparisson

Crowded Neighborhoods:

In ports and other crowded areas, bandwidths become congested with competing signals that interfere with one another. In addition, satellites and ground-based receivers can only take in so much information at one time. Any individual vessel may drop on and off the map because of this congestion. More satellites receiving AIS signals will help improve the coverage, but at the moment, we are limited by receiving capacity.

Drop Outs:

AIS signals can be spotty, and long gaps in transmission are not uncommon. There are a number of reasons for this, but the gaps to watch are ones in which a vessel captain turns his or her signal off to avoid detection. In these cases, we don’t know what they’re doing, but since AIS is a safety feature that helps captains avoid collisions with other ships, they eventually turn it back on. If a vessel seems to disappear in one location and reappear on the other side of a marine protected area, all we can say is: That’s a vessel to keep an eye on because they may have been fishing illegally. Read more about how our analysts handle gaps in AIS signals.

False Location:

Occasionally, tracks appear in impossible locations, say, over mountain ranges or through deserts. In such cases, either the AIS transponder has malfunctioned, or it has been tampered with deliberately. It’s our job to correct the position error and determine the vessel’s true location. See how our analysts are overcoming this challenge.

Mistaken Identity: 

Vessel operators have to input codes to their AIS by hand, which means errors are not uncommon. Again, it’s either a mistake, or a willful attempt at evading authorities. The codes include such information as the vessel’s name, type of boat, activity and a nine-digit identification number (known as MMSI for Maritime Mobile Service Identity). In many cases, information is left out or is incomplete. In some cases, operators deliberately enter false MMSI numbers. Other times, they enter only the first three numbers of their MMSI, indicating their country of origin, followed by zeros. Not only does this make it difficult to identify the vessel, but when multiple vessels from the same country do the same thing, it looks like there’s just one boat broadcasting the same MMSI from multiple locations. The resulting track shows a single vessel jumping across the ocean at impossible speeds. Read more about how our analysts are overcoming this challenge.

What Does Fishing Look Like Anyway?

When all the above challenges are taken care of, we still have to parse out which tracks among all the possible meanderings of ships on the sea represent fishing vessels, and when they’re actually engaged in fishing. That gets us back to the more than 127,000 vessels we saw broadcasting over a random 24-hour period. Of those, perhaps 30,000 are fishing vessels, and we’ve got to determine which ones they are. That amounts to a lot of vessel tracks to be evaluated and monitored.

Sample Vessel Tracks

We know purse seiners travel in a tight circle when they fish, enclosing a school in their nets. Longliners traverse an area back and forth as they alternately set their hooks and return to pull them in. Trawlers move at a constant speed dragging their gear behind them. Some may appear to zigzag as they change directions between trawls. But different size ships behave differently, and their tracks can vary. What’s more, a fishing vessel merely searching for fish may be hard to distinguish from one that’s actually setting and hauling their fishing gear. A non-fishing vessel may also appear to be fishing, for example when it’s actually making supply runs to and from another ship or oil rig. Read about how our research partners are helping us identify fishing behavior, and how we’re using machine learning to develop the Global Fishing Watch tool.

These are just some of the challenges we face as we develop Global Fishing Watch’s algorithms. There are many more. Today’s technology gives us the power to process millions of satellite signals at once (an impractical task for a staff of live people). But computers lack human intuition. They rely only on what they are told to look for, what they are told to ignore, and how they are told to interpret different bits of information.

It is only through a great deal of knowledge about fishing, vessel behavior, satellite signals and computer technology that our expert engineers and scientific partners are able to identify which key pieces of information a computer needs in order to understand false signals, spoofed tracks, mistaken identity and many, many other complicating factors.

The Global Fishing Watch algorithms we are developing are complex and robust. They are being built to handle the fine-scale nuances embedded in what initially seems like a pretty straightforward task: identify tens of thousands of fishing vessels out of some 127,000 vessels plying roughly 140 million square miles of ocean every day and determine what kind of fishing they’re engaged in as well as when, where, and how much of it they’re doing.

Scroll to Top