Go to Ghost Boat
Ghost Boat
Letter sent on Dec 13, 2016

What We Found At Sea

Trying to track the Ghost Boat—thanks to thousands of readers.

Earlier this year you helped us undertake a gargantuan task in the Ghost Boat investigation — searching more than 3,000 square miles of archived satellite photos of the Mediterranean Sea, looking for evidence of missing refugees.

Through a partnership with Tomnod and Digital Globe, more than 75,000 readers took part in this hunt, covering every inch of the search area. The data we gathered was overwhelming — more than 200,000 data points, more than 2,000 potential boat sightings.

It was so overwhelming, in fact, that we knew we needed help. So we went to one of the true experts: Michal Migurski, VP of product at Mapzen.

Mike is a mapping veteran, and looked at the data we’d collected and tried to make sense of it. He started off by plotting sightings that had been flagged up by readers. The early results were… a mess.

It was clear that the search areas had concentrated around two likely locations — Zuwarah in the west of our zone of interest, and Tripoli, the Libyan capital, to the east. But we also saw a handful of reports around Zawiyah, a city in the center of our search region. In any case, there were so many potential sightings that it was impossible to make much sense of things.

The next step was to see if we could filter out the noise. We did this by trying to connect together different sightings made at different moments in time; after all, the satellite images weren’t all taken at the same moment.

Taking into account that the Ghost Boat’s engine most likely was not working — after all, if it had a functioning one it probably wouldn’t have disappeared — we thought we might be able to spot sightings which could conceivably be a single boat drifting over time. Ocean currents are a complex mixture of wind and wave speed; but for this pass through, we took 5mph as the maximum speed that it might have drifted.

When we plotted connections over time, the maps looked like this:

Again, it was overwhelming. Around Tripoli, in particular, the density of potential connections was just too much.

As Mike said: “They all turned out to be mutually reachable.”


Instead of trying to run endless computations on a huge amount of data, we decided to try and narrow the field of results manually. Bobbie Johnson and Eric Reidy went through the thousands of sightings and looked at the tiles to see what had been spotted, focusing particularly on the first 10 days after the refugees’ disappearance.

They ranged from false positives — sightings that were definitely boats, but certainly not the boats we were looking for — to the simply unlikely (boats in and around ports), to those which had greater potential.

This filter dramatically reduced the number of possible sightings, down from almost 3,000 to just 165. That meant we could plot them again, so Mike got to work.

“I looked at every pair of scored images and determine how far apart they were in space and time,” he explains. “If they seemed to be close enough to one another to potentially be the same boat, I treated them as a pair and drew them on the map.”

“The brightness of the line shows the score that people gave each image — how reliable a sighting it was, and then I overlaid it on a Landsat aerial image from summer 2014, when the boat disappeared.”

Here’s the new map. You can see it in full resolution here.

The images, with their spidery lines and possible points of connection, are eerily beautiful. Still, the center of Tripoli and Zuwarah are condensed and bright, suggesting boats coming to and from port — or at least vessels in high-trafficked areas that many others would have passed through. Even if a boat was drifting in this zone, it would have most likely been spotted by somebody else.

On the fringes, though, were some sources potentially worth examining. Take the area to the east of Tripoli; isolated sightings that, if they were truly connected, behaved a little strangely.

But overall we couldn’t see obvious or specific patterns. For the most part, either the craft in each potentially-connected image was clearly different, or the potential sighting in one place simply didn’t connect to other plotted points.

We knew we could do more digging, but we were aware how fragile the exercise was getting. It’s not that we were searching for a needle in a haystack — it’s that we had no indication that the needle was even there. There was, after all, no guarantee that the boat was in this area, and that even if it was, it was ever captured by satellite.

And then… then we spoke to Measho Tesfamariam, one of the last people to see the Ghost Boat.

And suddenly things started to make sense.

Coming tomorrow: Episode 9—an interview with a convicted smuggler.