When two ships meet to transfer goods, it is called transshipment. In the fisheries industry, it is sometimes legal in ports, but usually illegal out at sea where the practice can’t be monitored. [You can read more about it here]. Transshipment can facilitate the mixing of illegal or unreported catch with legal catch, making it easier for illegal operators to “launder” their product.
Our goal is to be able to train the Global Fishing Watch model to identify transshipments based on patterns of vessel movement. The first step is to analyze AIS data from a lot of examples of verified transshipments. The challenge is finding those verified examples, which is what our analysts have been working on.
“We are trying to figure out the characteristics of transshipments that you can observe from the data that we have,” says our analyst, Nate Miller. We know, for instance, that when vessels transship, they sidle up next to one another and drift at very slow speed. We also know it can take hours, even days to transfer thousands of pounds of fish between vessels.
So, Nate and our team ran the AIS signals in our database through a crude filter to find instances in which two vessels were close together, moving slowly, for a minimum length of time. We’re calling these meet-ups “encounters.”
After experimenting with a number of combinations between vessel speed, time spent together and distance from one another, Nate has narrowed the criteria to define an encounter as 2 vessels within 500 meters of one another for 3 hours or more, traveling at a median speed of less than 2 knots during the encounter.
We use a distance of 500 meters because satellites may not capture the signal from both vessels during the time they are side-by side. We restrict the data to include encounters for which there are at least 20 data points (AIS signals), and we filter out encounters that occur within 20 nautical miles of shore to ensure that we’re not picking up vessels that meet in port.
Not all encounters represent transshipments, however, because there are many reasons two vessels may meet at sea—think of a tug for instance, or a supply ship bringing food, fuel and crew. So Nate is only looking at encounters between two vessels in which one is a refrigerated cargo vessel. Known as “reefers,” these ships do not fish, rather they receive catch from multiple fishing vessels and store it in enormous holds until they have enough to bring into port. We consider encounters with reefers to be possible transshipments.
The video below shows an encounter between two fishing vessels and a reefer off the east coast of Africa. The reefer track is shown in yellow.
Nate has also been analyzing data from reefer vessels alone to identify possible transshipment behavior in their tracks. That’s because large reefers that are required to transmit their location may be encountering small fishing vessels that are below the size requirement for AIS useage.
Refining the parameters that identify actual transshipments is still an ongoing process. Nate and the team have found multiple ways to evaluate the data to see what it can tell us. They are beginning to look at vessel tracks two days before through two days after an encounter to see if that reveals yet another set of characteristics that identifies possible transshipment.
Another look at the encounter captured above reveals that the two vessels are engaged in fishing before heading directly for a the rendezvous with the reefer, which is on a direct path for the same location.
Perhaps one of the most rewarding accomplishments will be to “ground-truth” our model by cross referencing output with a list of actual transshipment reports from the Indian Ocean Tuna Commission (IOTC). Aligning their list to our data is not a simple matter. Boats drift as they transship and AIS messages are not always updated minute by minute, so the point at which a vessel chooses to report transshipment to the IOTC may not match the point at which we receive an AIS signal.
To determine the best time window and distance radius to find matches, he is looking for transshipments in our data that occur within a latitude/longitude radius of 30 degrees two days before or two days after an IOTC-reported transshipment. We’re still working on the best combination, but some of the encounters are clearly matches—verifiable events we will be able to use to train our computer models.
“The ultimate goal would be to build an algorithm that could develop voyage reports describing all the port stops a reefer made over a given period of time,” says Nate, “what vessels it encountered or how many transshipment-like movements it made between port stops, and how likely these events are to have been transshipments.” This type of report would be a very useful, tangible application for the Global Fishing Watch algorithm. It would go a long way toward helping us identify hot-spots of potential transshipment and understand the extent of illegal and unreported catch flowing into the global seafood market.