The coastline you need is older than the radar
A community remembers a shoreline from the 1990s. Free SAR does not exist before 2014. Which instrument answers the question is not a preference, and the honest answer changes what you can promise.
· 7 minute read · Caelus
There is a particular kind of request that arrives from communities on low-lying coasts, and it is almost always some version of the same sentence. The water used to be further out. We know it. We have been saying it for thirty years. Can you show it?
It is a good question and it has a specific technical answer, which is that the instrument Caelus is built around cannot do it, and saying so early is more useful than discovering it three weeks in.
The archive problem
Free, systematic, globally available synthetic aperture radar starts with Sentinel-1A in 2014. There were earlier radar missions and some of their data is accessible, but as a dense regular time series you can difference year against year, 2014 is the floor.
A community describing change since the 1990s is asking about a period where the radar record does not exist. No amount of processing recovers a measurement nobody took.
Optical imagery is worse at almost everything this work needs, and it has one decisive advantage: it was already there.
Landsat 5 launched in 1984 and kept going for twenty-eight years. Landsat 7 from 1999, Landsat 8 from 2013, Landsat 9 from 2021. Thirty metre resolution, free, and a continuous record of most of the world's coastlines going back four decades. Sentinel-2 adds ten metre optical from 2015 onward for the recent end.
Thirty metres sounds coarse next to what modern radar resolves, and for a single date it is. For a trend it is far better than it sounds. You are not measuring one shoreline to the metre. You are fitting a line through dozens of observations of the same shoreline across decades, and the fit is considerably tighter than any individual measurement in it.
Two different problems wearing the same clothes
It helps to separate what is actually being asked, because coastal change requests usually contain two questions that need different instruments.
| Slow onset | Rapid onset | |
|---|---|---|
| Example | A shoreline retreating a metre or two a year over decades | A Category 5 cyclone rewriting a coast in one night |
| What resolves it | A long optical time series. Depth of archive beats resolution. | Before and after pairs, close together. Radar if the event is recent, because cloud. |
| Main difficulty | Tide. Every scene is at a different tide stage, and uncorrected that noise is larger than the trend. | Getting a clean before. The archive has to contain a usable scene from shortly beforehand. |
| What you can honestly say | A rate, with a confidence interval, and where along the coast it is fastest. | What changed between two dates. Attribution to the storm is inference, not measurement. |
The tide correction is the actual work
This is the part that gets skipped and it is the part that decides whether the number means anything.
A satellite passes at a fixed local time, but the tide does not care about the satellite's schedule. On a coast with a two metre tidal range and a shallow gradient, the waterline can sit a hundred metres apart between two scenes taken at the same place in the same year, with no erosion whatsoever. Compare an unlucky pair and you can report dramatic retreat, or dramatic accretion, entirely from tide stage.
Which means the pipeline is not extract waterline, difference, publish. It is extract waterline, pair every scene with a modelled tide height at acquisition time, normalise to a common datum, and only then fit the trend. Scenes that cannot be reliably corrected get dropped rather than included with a caveat.
Why this is worth the trouble
Caelus is in early conversations with PISFCC, the Pacific Islands students' organisation that took a climate advisory opinion to the International Court of Justice and won it in 2025. They collect oral testimony from frontline communities: what people saw, what they lost, what the coast used to be.
Testimony carries what no instrument holds, which is what the change did to the people living through it. What it does not carry is a number, and in the rooms where obligation and resource allocation are argued, an account without a measurement beside it is easier to set aside. Not because it is less true. Because those rooms are built to discount it.
The three Vanuatu sites under discussion split cleanly along the line drawn above. Blacksand and Eton on Efate are slow-onset shoreline retreat, which is the tractable case: forty years of Landsat, tide corrected, gives a rate and an interval. Worasiviu on Pele Island is the same class of problem, with the added complication that the community has already built a sea wall there, so recent shoreline position reflects an intervention rather than only the process. Niue is the rapid-onset case: Cyclone Heta came through as a Category 5 on 5 January 2004 and reshaped the north and west coasts. Twenty-two years back, before the radar record, which puts it squarely in the optical archive and makes the useful output a trend and a projection rather than a damage map for an event everybody already remembers.
Where this came from.
Credited in plain text, never as logos. Use of open data is not a relationship, and a logo would imply one.
- Landsat programme
- USGS and NASA. Continuous optical record from 1984.
- Sentinel-2
- ESA. Ten metre optical from 2015.
- Sentinel-1A launch, 2014
- ESA. The start of the free systematic SAR archive.
- Cyclone Heta, 5 January 2004
- NIWA Island Climate Update. Category 5, with the centre passing west of Niue near a spring high tide.