For non-serious fields, game-terrain editing to be specific, I've got to generate much more detailed DEM (5-meter at least) from 1-arc-sec SRTM, which has the best resolution I can find yet.

I wonder if there's any interpolation / noise-generation / etc. algorithm that can re-sample lower-resolution DEM and create details accordingly without harming the credibility of the terrain to human eyes.

  • 1 arcsecond SRTM (I assume you mean) isn't too bad but you can't create detail that isn't there. There are various resampling methods that you can employ dependent on your software but most of these will smooth rather than sharpen the elevations. There will most likely be better elevation models of most areas available but they will not be 'free' like SRTM. – Michael Stimson Mar 1 '17 at 1:40
  • @MichaelMiles-Stimson Thank you for typo-correction! Popular re-sampling method smoothing elevations is just the point that besets me. – Schezuk Mar 3 '17 at 5:01

If you have access to Spatial Analyst in ArcGIS (or something similar), you can do something like the following:

If you want to generate 'noise' to your DEM, I suggest using the Create Random Raster tool to generate a random raster at the same extent as your SRTM DEM. You can specify the output cell size to 5m or whatever you require. The results of the pixels will be between 0 and 1.0. You can add this to your DEM using the Plus tool ensuring your cell size in the environments is set to the same as your random raster.

If you need to, before you add the DEM and random raster, you can multiply the random raster by a constant using the Times tool. For example, multiplying the random raster by 2 would result in values between 0 and 2.0, which you could then add to your DEM.

That should create sufficient detail that doesn't modify the original DEM too much.

  • To composite terrains with high-res random raster may not generate convincing details. For example, to tessellate a gully in low-res DEM, rather than random patterns, ditches paralleling to the gully should be applied. I guess high-res DEM (e.g. with 5-meter cells) would demonstrate similar fractal features at a smaller scale as low-res DEM (e.g. with 30-meter cells) would do in each spot, which should be considered in the re-sampling algorithm to select. – Schezuk Mar 3 '17 at 5:43

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.