2

I'm generating elevation contours from the latest ASTER DEMs, which involves processing hundreds of tiles. Unfortunately, some of them have substantial errors, as seen in this colour relief from S01W079:

sinks

These patches are not voids. They have "valid" elevations that are just wildly inaccurate. The resulting contours (generated with gdal_contour) look like this:

contours

I need to find these spots, but visually inspecting every single tile for errors is not an option. Is there any method I can use to automatically identify problem tiles? CLI and Pythons solutions preferred, but anything would help.

Right now the only option I see is to write a Python script that picks up strange jumps or drops in elevation across the DEM.

4

If you want to pick them manually doing it on a shaded relief visualization helps. The best I've found is to make a mosaic of my whole area of interest and take the difference with another DEM. SRTM for example. The the errors pop up as areas were the difference id very large and are very easy to spot. However, you have to first check that the large difference doesn't come from errors in your reference DEM.

Also, if by the "latest ASTER DEMs" you mean the ASTER GDEM, you can use the coverage data, and directly mark as invalid areas all the ones with no coverage, that will save you a lot of work. However, there are errors also in areas with small coverage values.

And example of the first approach. This is a section of the GDEM ASTER GDEM

And this is the same section on the SRTM enter image description here

Large differences are not obvious, but take the unsigned difference and only show values over, let's say 50 m, this show up enter image description here

And the large differences are pretty easy to spot there.

Be aware that SRTM has voids that have been patched with GDEM, so it is better to use a non-void-filled version of the SRTM.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.