[OSM-dev] Problem with Postgresql / Mapnik / SRTM Contours

[Andy Allan]

On Mon, Nov 24, 2008 at 12:25 PM, sylvain letuffe <liste at letuffe.org> wrote:

> Looks like someone of the opencyclemap (is that you andy ?) have included some
> sort of interpolation or any other means to replace those old "white zones"
> by "at least something"

Sylvain wins the prize for noticing!

Yep, it's known in the trade as "void-filling", if you want to read
back through some of the mailing list archives on the subject. It's
not specifically snow-covered areas, it's areas with no data from the
SRTM mission - often found in high mountains (since the shuttle can't
see especially steep north slopes, for example) but generally in areas
with no radar return (including marshland).

> See here what I'm talking about :
> http://www.opencyclemap.org/?zoom=16&lat=45.3086&lon=5.85346&layers=B000
>
> The POI "dent de crolles" is the highest point and you see the contours
> being "very" incorect

The void-filling is currently a pattern-based interpolation, so the
areas of voids are even more fictional than the rest of them!

> As I am on the process of constructing an hiking map, I am searching for
> solutions... which might not even exist.
>
> I first suppose that there are not "free data" for contours at all (exept free
> for personnal use SRTM data) am I right ?
>
> Has anyone tried other sources ? what about SRTM v3 ? is there a wiki page
> that gather those research ?

There's no SRTM v3.
v1 - contours including noise from oceans and other water bodies.
Nobody uses it.
v2 - contours 'clipped' to water bodies
SRTM1 - high res, USA only
SRTM3 - medium res, world wide

So I use SRTM3 v2, and it's the same for the piste map and so on. Both
SRTM1 and SRTM3 have the same voids, since they come from the same
ultimate data source.

As for void filling, that's a huge project with lots of tradeoffs. It
basically comes down to synthesising data, usually with a sprinkling
of other data sources. So the algorithm used to void-fill the cycle
map could have added constraints to it like making that mountain the
right spot height and guessing the rest of the gaps. The most
sophisticated models also do things like ensuring rivers run downwards
(hydrological models). But it's a very hard thing to do on a global
scale without using many man hours per void.

> Has anyone worked on using GPS tracks elevation information to improve SRTM ?

GPS elevation tracks aren't better than SRTM in general, due to their
poor vertical resolution. Whilst they could be used for void-filling,
it's generally only the position of the summits that could help - most
voids are in inaccessible areas and the chances of getting sufficient
coverage are remote. Better solutions involve other data source, e.g.
other maps with compatible licenses where the contours could be
digitised.

I've been asked a few times as to how to "crowd source" void filling,
and the best answers I have are summit heights and river courses.
Beyond those two constraints for the model, better sources are
required that generally are centrally gathered.

Cheers,
Andy