← back
2026CNG Japan 2026 speaker JAMSTEC Yokohama talk
interactive · cng.surreal.red · overview

The surveyors I work with want their data used — and they told me there still aren't many use cases. So the question I started from: how does a delivered dataset become something people keep using?

Kofu City's point cloud is a typical case. 2,008 tiles on the 1/500 図郭 grid cover the city, each 400 × 300 m — one ZIP per tile, per product, around 200 MB each. The data is open, properly licensed, and carefully made. But getting into it means learning a stack of tools. I'm a designer — I couldn't really use it either. People use what's easy to use, so fewer steps should mean more use cases.

So, an experiment: one city, two steps. Convert it into something a browser can open, then see whether use cases actually turn up. Kofu is 212 km², about 9.6 billion points in the source — and the result is one 13.9 GB COPC file and two COG terrain files. No downloads; just navigate. The pipeline tried EPT, Potree and 3D Tiles along the way, and the two formats that got published are exactly the two that are single files.

What does it cost to leave it up? 18.7 GB on object storage, $0.28 a month, zero egress — about $3 a year. And nothing gets replaced: the 図郭 tiles stay exactly where they are; this is one more object sitting next to them. The front door is a single STAC catalog URL, carrying the CRS, the extent, the licence, and a link back to where the data came from.

Two things I'd like to see, and neither one is big: for the data that already exists, one cloud-native copy alongside it; for the data made from here on, publish it this way from the start. The second is the easier of the two — easier to publish, easier to use, better for everyone. There are 68 more point-cloud datasets like this on the national portal; Yamanashi's aerial LiDAR alone is 37,926 files and 2.6 TB. If more of this data were published this way, I think a lot more people would use it.