Explore improvements to "best" flag

[tyrasd]

The best attribute to flag imagery sources that are the best layer to display by default in a given region doesn't always work well – for example when an imagery is only better than another imagery in a fraction of its coverage area (e.g. see #112 (comment) ff.).

Let's discuss possible improvements here.

[imagico]

For this project a useful quality metric in addition to the capture date would be ground resolution. This of course only can be specified for homogeneous image sources. But together with the date it would allow applications to perform a better assessment of the image sources. If for example a source like Bing reaches the limit as you zoom in or has an extremely old date specified in the tile metadata (like the standard 1/1/1999-12/31/2003 for the Geocover basemap) the application could look into the available other sources and suggest to the user to use whatever promises a better quality at the current location.

[simonpoole]

@imagico that would not address the issue that quality can vary drastically in the same imagery source even with the same nominal resolution (naturally we don't even record that currently) and in the end quality in this case is only meaningful as a relative measure (not to mention all the issues with zoom levels and so on) and that is why I would prefer a solution as I outline here #112 (comment)

[pnorman]

We have an indication of approximate resolution with maxzoom for TMS sources. It's only approximate, but imagery resolution differences are frequently several powers of 2.

[imagico]

Due to the variable scale of the Mercator projection the maximum zoom level does not say much about the resolution. In the osmim the maximum zoom currently varies between 10 and 14 for example despite all images currently offering either 10m or 15m ground resolution.

[pnorman]

Due to the variable scale of the Mercator projection the maximum zoom level does not say much about the resolution. In the osmim the maximum zoom currently varies between 10 and 14 for example despite all images currently offering either 10m or 15m ground resolution.

For imagery over a reasonably small area, like most, the variable scale doesn't matter as it impacts all imagery the same. A bigger current issue is probably the high number of sources with 22 as a max zoom (~2-3cm where it is) but the imagery is 20-50cm

[bhousel]

from: #164 (comment)

Our hope for this best flag is that local contributors can use it to mark imagery that they know to be better than Bing/Mapbox. For example, "Kanton Zürich 2015 10cm" is marked "best" because people from Switzerland say so, not because of the imagery's age and resolution. Tomorrow Bing may update their imagery and then "Kanton Zürich 2015 10cm" will no longer be the best source there, and ideally somebody from the Swiss community will notice this and open an issue to remove the flag.

Closing this issue for now.. Consider "best" as meaning "better than Bing according to local mappers".

[Klumbumbus]

Another "solution" might be to turn around the whole "best" concept:

Remove the best attribute from all imagery entries. Create new separate area polygons and assign one layer as best to this area (by the id of the layer). These areas can be a whole country or a smaller part or whatever.

However this doesn't solve all problems and creates new ones:

• "best" can still mean different things (date, clouds, resolution, orthorectification,...)
• the new polygons must not overlap. So a common database for all these polygons would be required. Something like the osm database just for these polygons.

Not sure if it is worth the effort. It all sounds a bit complex too.

[bhousel]

To be honest, I'd prefer to just build an entirely new thing that can keep track of - for a given quadkey which imagery is best, along with whatever adjustments need to be made (offsets, brighness/sharpness, etc).

@Zverik's imagery offset database is pretty good, but I'd like to take it a step further!

[bhousel]

oh also - I'm reopening this issue.. best was a good enough hack when we really only had Bing, Mapbox, and whatever local imagery to choose from. We have so many more imagery to trace from now that we need something a bit more nuanced (this is a great problem to have!)

[grischard]

What about having a bestoverride geojson featurecollection that has polygons where the best background is their bestmapid attribute? Or do we want something more objective than "best"?

[grischard]

There was also :
Best-best image for region? #390

[crackwitz]

I must also question the utility of a simple "best" attribute. This merely splits available imagery sources into "first class" and "second class", without quantitative meaning. This is especially noticeable near borders, where (1) the neighboring classifications spill over and (2) the neighbors show your area as blank.

I think the situation needs objective attributes quantifying the image data. The common points of contention (in the "best" debate) are vintage, resolution, and coverage.

• Vintage: some sources have it, some don't. The date ranges vary too, some stating a single month, some spanning a year or more. This should ideally be attributed to each tile, but if a polygon for the imagery source exists, then for that polygon.
• Resolution: hopefully a decent measure of actual optical resolution (what you can see), regardless of digital resolution (the pixels). If I lay a checkerboard pattern of black and white tiles on the ground, the tiles being 0.1 m along each edge, and the pattern is blurry or hard to make out (3 dB cutoff or whatever), then that should not be called "0.1 m resolution".
• Coverage: plenty of locally high resolution imagery sources have associated polygons. If any attribute (resolution, vintage) differs noticeably in this polygon, then why not split the polygon into several?

For many sources, those are hard to get authoritative values for. I think it's reasonable to guess date ranges and resolution from imagery, and to draw polygons by what the source actually shows right now.

With such attributes, users could choose their own ranking of available sources. Do I want the most recent images, regardless of resolution? Do I want the best resolution, even if it's older? Do I want some mixed weighting? Who else can tell but I?