What twenty-five years of this work actually looks like

I got on a call with John Amos expecting to talk about oil slicks. I came off it thinking about funding, mostly, and about how long twenty-five years actually is when you are the one living through them.

John founded SkyTruth in 2001. He was a geologist working in the private sector before that, and the thing that pulled him out of it was a NASA-funded project in the early nineties using radar to spot natural oil seeps from orbit. He has been at this, in one form or another, since before Sentinel-1 existed, since before satellite imagery was something a nonprofit could realistically build a product on rather than something only a government or an oil company could afford to task.

The funding conversation nobody has in public

I asked him directly what the hardest part has been, and the answer was not technical. It was money, and specifically the mismatch between how this kind of work has to be funded and how the people who fund it want to measure it.

A grant cycle wants a deliverable by next quarter. A twenty-five-year monitoring record only starts paying off after the fifth year, when the baseline is long enough to show a trend instead of a snapshot.

That is not a complaint about funders being short-sighted, or not only that. It is a structural mismatch between the timescale humanitarian and environmental monitoring actually runs on and the timescale most funding instruments are built to reward. A company can raise a round against a product roadmap. A monitoring archive has to raise, over and over, against the argument that the record it is building will matter more in year ten than it does today, which is a much harder pitch to make to someone who needs to report back this year.

We talked about Global Fishing Watch, which SkyTruth co-founded with Google and Oceana, as the example of what it looks like when that argument actually lands: a public, free vessel-tracking platform that took years of unglamorous grant funding before it became infrastructure other organisations build on top of. Nobody writes a check for infrastructure on day one. They write it once the infrastructure already half exists, which is exactly the chicken-and-egg problem every early-stage nonprofit in this space is fighting.

The Satellite Sentinel Project, as a cautionary tale

John brought up the Satellite Sentinel Project on his own, not as something SkyTruth ran, but as the clearest example either of us could point to of what happens when satellite monitoring gets a launch moment with real star power behind it. George Clooney and John Prendergast started it in October 2010, on a trip to the Sudan and South Sudan border, working with DigitalGlobe imagery to watch for a return to civil war and to document threats to civilians in near real time. Within its first year it had documented attacks, mass displacement, and mass graves. It was, by any measure, a serious and effective piece of monitoring, and it had a level of public attention that almost nothing in this field gets.

What we talked about was not whether it worked. It did. What we talked about was what it takes to keep something like that running once the launch moment passes and the news cycle moves on, which is the part celebrity attention does not solve by itself. Attention gets a project funded once. It does not by itself get a project funded for the fifteenth year, and the fifteenth year is usually where the actual value of a long monitoring record shows up.

Cerulean, and why the boring tool is the right one

The product SkyTruth actually ships now is Cerulean, which went publicly live in December 2023 and is, as far as I know, the first free public tool that automatically detects ocean oil pollution from satellite radar and tries to attribute it to a source. It runs as a fully automated pipeline over every Sentinel-1 pass: no analyst has to go looking for a slick, the model finds candidate slicks in the radar backscatter the same way a human interpreter would, and where there is vessel traffic data to check against, it tries to connect a slick to the ship or the infrastructure most likely responsible.

The number that stuck with me, which John has stated publicly, is that of roughly 90,000 slicks Cerulean has logged, only about half a percent were ever officially reported by anyone. That is the whole argument for the tool in one figure. The ocean does not lack slicks. It lacks a record of them that anyone was keeping before a machine started keeping it automatically, every pass, whether or not a person was watching that week.

The unglamorous version of a monitoring tool is usually the one that survives, because it does not depend on someone deciding to go look.

Where the same logic goes next

The part of the call I came away from most energised by was the parallel structure. Strip Cerulean down to its actual components and it is: an open radar archive, a detector trained to find a specific signature in that archive, a public record of every detection, and an attempt to attribute each one to a cause. None of those four steps is specific to oil. John pointed at flood extent and wildfire smoke as the two places he sees the same structure landing next, which is not a surprise to me, since it describes most of what Caelus is trying to do in Rasuwa and in the Niger Delta.

The difference is where the attribution question points. Cerulean's hardest problem is tying a slick to a specific vessel or platform. Ours is usually simpler on the attribution side, since a flood or a flare has an obvious physical cause, and harder on the delivery side, since the open question is not who caused it but who is close enough to it, and poor enough, and unrecorded enough, that nobody has told them yet.

We did not solve anything on that call. What I took from it was closer to reassurance than to a plan: that twenty-five years in, the founder of one of the organisations that made this kind of monitoring possible in the first place is still fighting the same funding mismatch we are fighting in year one, and is still betting on the unglamorous, automated, public version of the tool over the impressive one. That is a good bet to be copying.