[GRASS-user] wierd artifacts in DEM from seamless.usgs.gov

[Dylan Beaudette]

Thanks to everyone for the helpful advice. 

After a couple hours of working with the data here is what I found out:

data source: 1/3 arc second DEM from seamless.usgs.gov (lat/lon)

projected to utm via :
gdalwarp -tps
gdalwarp -tps -rb
gdalwarp -tps -rcs
r.proj
Arc/Info project grid CUBIC
r.to.vect -> v.proj -> v.surf.rst

graphical overview can be found here:

http://169.237.35.250/~dylan/gdalwarp/image_matrix.html

in terms of the shaded relief r.proj produced the best _looking_ image. It 
turned out that the original cross-hatch artifacts were a result of the 
default nearest neighbor routine used in gdalwarp. The source data does not 
appear to be afflicted with these artifacts, although their presence in all 
of the projected version suggests that there might be something wrong...

Vertical artifacts appear in all of the projected images, except for that of 
r.proj using the cubic convolution sampling method. (how exactly does r.proj 
work?). The best _looking_ tangential curvature raster was derived from 
gdalwarp with the -tps and -rvc (cubic resampling) flags. However, this is 
probably due to the inherent smoothing that goes along with this set of 
options- as Maciek pointed out. 

converting raster -> points -> v.proj -> v.surf.rst seemd to work ok (no 
artifacts in  the shaded relief besides edge effects), albeit very time 
consuming.

also worth noting is the severe contour-bias seen in the profile curvature 
maps.

Dylan

On Friday 19 May 2006 12:58, Maciek Sieczka wrote:
> On Thu, 18 May 2006 19:04:36 -0700
>
> Dylan Beaudette <dylan.beaudette at gmail.com> wrote:
> > Howdy,
> >
> > downloaded a 1/3 arc-second (about 10m res) DEM from the
> > seamless.usgs.gov site.
> >
> > projected the lat/lon raster to UTM via gdalwarp using the -tps option
> >
> > import into grass, and compute shaded relief... and it looks like
> > crap: http://169.237.35.250/~dylan/temp/seamless_dem_artifacts.png
>
> Dylan,
>
> When I need to reproject a DEM raster, first I convert it to points
> (r.to.vect), then reproject the vector and re-interpolate DEM out of
> it with v.surf.idw or v.surf.rst. This method yields lowest disturbance
> in DEM look and feel, derivatives (curvature, slope, aspect are those I
> tried) and preserves the min max best.
>
> Due to high memory demand of Grass >= 5.7 vector engine you might run
> into troubles though, then 5.4 and sites is your friend. Or transform
> raster into x,y,z list (r.stats), reproject with cs2cs, import as vector
> with v.in.ascii in points mode, interpolate this?
>
> The only way you could avoid such stripping artifacts in the
> reprojected DEM, without going trhough raster->vector->raster, is to use
> gdalwarp's cubic spline resampling (-rcs), as other resampling methods
> (bilinear, cubic) are not able to fully remove stripping artifacts. On
> the other hand, the amount of smoothing involved by -rcs is
> unacceptable for most applications besides rough visualisation IMHO.
>
> You might like to search for messages several months ago by Frank
> Warmerdam and Paul Kelly, either in Gdal list or one of the Grass lists,
> in reply to my questions about DEM reprojection.
>
> Maciek
>
> --------------------
> W polskim Internecie s? setki milion?w stron. My przekazujemy Tobie tylko
> najlepsze z nich! http://katalog.panoramainternetu.pl/

-- 
Dylan Beaudette
Soils and Biogeochemistry Graduate Group
University of California at Davis
530.754.7341