Elevation Contours and DEM experiment results

Hi all,

as we had the original discussion about elevation contours on the
flightgear-devel list, I'm crossposting this to flightgear-devel and to
terragear-devel, however I'd suggest replying to terragear-devel as this
is essentially a TerraGear issue.

After the discussion about integrating elevation contours into
TerraGear-generated scenery I did some experiments using digitised
elevation contours from current topographical maps and interpolations
from the standard SRTM DEM data.

I used GRASS to digitise contour data from scanned topographical maps
and generate a DEM model with about 4m/pixel resolution for a small
region using the v.surf.rst tool from GRASS. The resolution is quite
arbitrary, as it is the resolution of my scanned topomaps.

Then I interpolated the SRTM data for the same region and resolution
using a cubic interpolation. I found that both the interpolated SRTM
DEMs and the DEMs generated from the contour lines are quite similar.

Generating contours from the interpolated SRTM DEMs and laying them over
the digitised contours confirmed this finding. The contours do not
differ substantially from each other.

I then interpolated the SRTM DEMs for some other regions, generated
contours and overlaid them to the topomaps. Again, the contours from the
interpolated SRTM DEMs quite well match those from the topomaps - at
least I think that the differences in terrain won't be noticable for the
human eye.

Unfortunately I've thrown away the manually digitised countour lines -
being a bit frustrated by the result - and for copyright reasons I
cannot post an overlay of the contours generated from the SRTM DEMs and
the topomaps I'm using. I'll however repeat the digitalisation for
another area and post that together with the respective SRTM DEM
contours if anybody is interested in a direct comparison.

I then tried to import the fitted irregular elevation grid generated by
Terra into GRASS and created an appropriate DEM for a small region.
v.surf.rst does a spline interpolation, which will differ substantially
from the flat approximation done by Terra and FlightGear for obvious 
reasons. I haven't found an  appropriate tool in GRASS to do a linear 
interpolation, however I tried v.surf.rst with a smoothing setting of 0, 
making the interpolated surface go directly through the points of the 
irregular grid instead of approximating them smoothly.

I calculated the differences between this interpolation and the
interpolated SRTM DEM and found a standard deviation of about 43m. The
irregular grid was generated using a maximum error of 40m and a maximum
number of 1000 nodes. However the maximum deviation between the two DEMs
is about 240m at at least one point. However this could be due to the
non-linear interpolation of the grid.

My conclusions from this are twofold:

1. There is no need to digitise contour data manually instead of 
extracting them from SRTM if your map material is based on SRTM - which 
seems to be the case for all of the maps I'm using.
2. The only "advantage" of integrating contour lines directly into the 
terrain would be to effectively bypass the fitting done by Terra.

I'll try to get some comparisons up with real-life photos and 
Terra-fitted terrain, but I already got the impression that at least in 
the area we're working on with our custom scenery the hills and 
mountains just don't look right ;-)

Maybe applying an interpolation in Terra similar to the one done by 
GRASS could yield some better looking results. Just thinking out loud here.

Regards,
Ralf


_______________________________________________
Terragear-devel mailing list
[email protected]
http://mail.flightgear.org/mailman/listinfo/terragear-devel
2f585eeea02e2c79d7b1d8c4963bae2d