Hacker News Reader

Comment by bluenose69 | Hacker News Reader

This is an intriguing site, part of a series that is also intriguing. One thing I hadn't noticed before (see https://solutionspace.blog/2022/05/13/map-data-i-its-all-abo...) is that openstreetmap gives a lot of digits on coordinates. I see 45.6684436 listed as a latitude. If that last digit were changed to a 7, the location would shift by about 1cm. In lots of places, plate tectonics is moving things by about that amount per year.

I wonder whether openstreetmap is automatically shifting things in its database, using a tectonics model.

I also wonder why they are giving so many digits for user-entered data. Can a person buy an affordable GPS that gives centimetre resolution?

View on HN

Coordinates in cartographic systems are traditionally fixed point, decimal or binary depending on the system. Many sources are internally quantized to meter resolution, though some newer ones are quantized to centimeter resolution. Because so much of this data is presented as floating point, it introduces a lot of apparent false precision.

The limits of practical precision for a fixed point on Earth is around 10 cm regardless of the measurement technique. Anything more precise than that is generally not reproducible. Fixed points on Earth will often quasi-randomly deviate from their median position by multiple centimeters over periods of hours regardless of the reference frame. In modern geospatial systems, it is not uncommon to have 1-cm defined as the physics floor such that any additional implied positioning precision can be disregarded or discarded for computational purposes.

OSM internally stores lat/lon in integer columns in a Postgres db, where the stored value is lat*10000000 (and lon*10000000). There is no tectonics model applied.

Somebody noticed (https://www.openstreetmap.org/user/StephaneP/diary/390290) that OSM isn't even consistent about which datum it uses. (Discussed in HN at the time: https://news.ycombinator.com/item?id=20460596 ). (Though to be fair, the datum error is probably smaller than e.g. GPS error in most cases.)

As for the "why" bit, the answer is probably "store and report whatever the user reports".

I don't think OSM data are actually that precise.

Also - I think moving data with some model is highly impractical. Location shift of 1cm is very small - things can move this small for various reasons - erosion, wear of material or construction nearby.

Yes. Real-time Kinematic Positioning (RTK)[0] is a differential GNSS technique of combining data from satellites and fixed ground stations. Since the location of fixed ground stations has been surveyed, and has a higher degree of accuracy, combining the data can provide a higher degree of accuracy to the downstream consumer of the data.

Some states in the US provide RTK data over the internet for free as a public service.

If your GPS supports RTK[1], and you have an internet connection, you can feed the publicly available RTK data into the GPS to get very accurate location data.

I've successfully integrated RTK and seen ~1cm accuracy. Our navigation device was not cheap, but I think the GNSS chipset in it was pretty common. To get really good GNSS accuracy you will need a clear view of the sky, as you'll need good satellite coverage in all directions to start with. You'll need an antenna mounted in such a way that it's not obstructed. Your antenna will need to be electrically grounded/isolated in such a way that noise is not introduced to the signal before it gets to the GNSS device. Etc.

[0] - https://en.wikipedia.org/wiki/Real-time_kinematic_positionin...

[1] - https://learn.sparkfun.com/tutorials/what-is-gps-rtk/all

It is too precise for an absolute global position.

But for relative posistions, i.e. the difference between to positions, 1 cm is about right. It is close to the resolution in the best aerial images.

So if you you want to map e.g., a park bench as a 2-dimensional object og the layout of railway tracks you need the precision.

When you zoom in a lot on a digital map you still want the details even if it is all off by several meters.

Just as you can have detailed printed maps of a small area.

There's no shifting going on, what users put in is what OSM stores, nothing more, nothing less.

As to why, I wasn't in the room when that decision was made, but the design philosophy of the OSM data model is to leave things as open-ended as possible while still being practical. An object's attributes can be any arbitrary combination of key/value pairs. There's conventions, but no enforced standards. My best guess is that the precision for longitude / latitude is an outcome of that same design philosophy. Back in 2004 when OSM started and all of this was defined, there was no way <10cm coordinate precision was going to be meaningful, but now you see use cases where it may be, like indoor mapping (see for example https://openlevelup.net/?l=0#19/48.87598/2.32630)

> I wonder whether openstreetmap is automatically shifting things in its database, using a tectonics model.

I’d be surprised if it did. Most of the data — both GPS and tracing aerial photography — is nowhere near the single-cm precision level to start with.

Users might enter data from different sources than their own GPS. I don't think that 1cm resolution is possible with GPS (the signals can have systematic errors for example caused by reflections).

https://xkcd.com/2170/

Reply on news.ycombinator.com