Digital Elevation Data
The term "DEM" (Digital Elevation Model) can refer either to a specific elevation
file format, or to gridded elevation data in general. Some people use
"DTM" (Digital Terrain Model) as the more general term.
File Formats
Regular Grid Quality Issues
- DEMs suffer from a large number of large number of problems, including gridding
artifacts, contour artifacts, and mismatches with other data sources
- "sinks" or "pits"
- for example, if you try to drape a river on a DEM, you may find that
the river must run uphill!
- things like lakes and depressions are valid sinks but when a sink develops
where a stream should obviously flow, its a problem.
- in some cases you can improve a DEM by enforcing the constraints of other
data (e.g. hydro)
-
Quantifying
and visualizing terrain fabric from digital elevation models, Peter Guth
- The Geomorphological Characterisation
of Digital Elevation Models, Joseph Wood
- describes a number of DEM problems: Integer Rounding Interpolation
Terracing, Terrain Flattening, Planimetric Offset
- Suzanne P. Wechsler's dissertation
DEM Uncertainty: Evaluation and Effect on Topographic Parameters, now hosted
on ESRI's site
Interpolation and Sampling
Compressing Elevation Data
- it is not recommended to apply conventional lossy compression techniques
to elevation grids
- Bryan Turner says:
- "All of the lossy formats I've tried (JPEG, IFS/Fractal, Wavelet) produce
some amount of distortion that is recognizable in 3D. Of those, Wavelets
were the least problematic and JPEG was the worst. IFS could produce excellent
results, or a mass of random garbage, depending on the data and the compression
rate. JPEG consistently produced flattened square patches with steep dropoffs
at the edge. Wavelets produced aliased versions of the heightmap at each
level, but usually kept the overall shape of the terrain. In stress testing,
it often flattened, or 'blurred' the terrain by eliminating steep areas
and filling in crevices, rather than keeping their relative heights intact."
- "All of the lossless formats I've used (Huffman, Adaptive Huffman, Level
3 Stat Model + Arithmetic Coding, LZ77, and LZW) are undesirable for paging
unless they are used separately on a block. Otherwise finding the correct
stretch of area to uncompress becomes a nightmare. I tried using them in-memory
as well, but reading a compressed format on the fly is too slow."
Cell Registration
- is it good to know the cell registration
of an elevation grid, though nearly all elevation grids have data which is cell-centered