Building DEM of large areas

Question

I am putting together a compilation of 1/3 arc second, and 1/9 arc second DEM data for essentially all of southeastern Oregon in ArcGIS. This surface would be used for geomorphological and geological modelling of various types, and to show areas and points where previous archaeological and geological research has been done.

The question is: what are the best practices one would use to put together such a large mosaic of high resolution DEM data? Is there any recommended literature I should check out? Optimally, this would all be in one map, but I worry about it being unwieldy and taxing on memory, especially when analyses are being run. Perhaps it should be broken down into subregions of some type?

Answer 1

Depending on your image type you can use Esri Raster Mosaic dataset - have your tiles and mosaic too without any (much) extra storage.

The mosaic dataset works like a VRT file where the rasters are referenced but not imported so the table in the database stores the path and location of the tiles; overviews are a good idea but they don't take up much space. For the purposes of geoprocessing you can select the tile or the mosaic dataset; all raster geoprocessing tools will use the mosaic dataset as a single raster.

As for actually creating it, consider a Terrain which you can load all sorts of elevation source data, even to a very large size. Then using your output extent setting on your tools (and snap raster as well, just to keep it neat) you can export tiles, more than one at a time, as raster.

For GIS software that's not Esri you can also have the same tiles as a VRT raster, make it using GDALBuildVRT. This is also an Esri raster type but is significantly slower than a mosaic dataset (notably, many times slower).

Hillshades help visualize terrain, which can be created in Esri using spatial analyst. Should you not have spatial analyst or not like the Esri hillshade (it's a visual preference) then you can use GDALDEM to create one, or tiles and mosaic that too, the DEM must be in a GDAL format (sorry, mosaic dataset isn't a GDAL format). You can use a VRT of the tiles, however this format has proven to be very slow on large datasets, best to use ERDAS Imagine (.img) or GeoTIFF tiles (with some overlap) and hillshade each one then clip the overlap area (GDAL_Translate -projwin or Esri Clip_management) and build a hillshade mosaic dataset.

If you aren't comfortable with command line, you can do a hillshade from QGIS in the geotoolbox:

Answer 2

I've done some work of this manner. Archaeologists love their giant DEMs for some reason...

I don't have anything formal but can lend some insight from experience. I found it was nice to have something that covered the whole area but it doesn't need to be at the highest resolution you've got. For example, a mosaic of the DEMs for the whole area at 1/3 arc second (or smaller) is nice to have. It's less taxing on resources and you can make locator maps of your whole data set with it.

I always like to keep the higher resolution stuff as tiles (existing 24k map boundaries work nicely) rather than stitch them all together. That way you can pick and choose the ones that are most relevant to whatever you're looking at for a given question.

I once had a problem where several adjacent DEMs used different units so make sure to check that out to avoid really weird cliffs on your tile boundaries.

I always like to rename the tiles to coincide with the quad map they relate to, it always makes it easier to find when you've got a long list.

Lastly, just as a matter of opinion, I think that hillshades look really nice set to resample the display using bilinear interpolation and about 20 percent transparency.