Accuracy of DEMs

Hello,

What is the best numerical way to measure the accuracy of a Digital
Elevation Model (DEM) if I don't have a reference map of the site? I
produce the DEM from an input file filled with measurements after
applying some filtering.

Thanks.

On Sun, 9 Nov 2008, Bumbala wrote:


My suggestion would be to reserve a number of control points for
cross-validation. Assuming your survey data is dense-enough to produce a
quality regular grid (cm to dm-spacing of grid cell points), it's easy
enough to compute the DEM surface, then compare the predicted elevation to
the elevation observed at the control points. You should then have a
distribution of lengths by which your surface deviates from the data; you
can apply whatever probability distribution you like to the errors for use
in prediction (Central Limit Theorem can be used to say that the errors
will likely be normally-distributed if they are random, but your mileage
may vary).

Why not use the control points in the generation of the DEM? If you use an
interpolation method that is constrained to go through all points (i.e.
an exact estimator), you'll get out what you put it. So you should reserve
at least a few control points for checking the DEM output.

Good references:
http://edc.usgs.gov/guides/dem.html

Regardless of your method, you still need 'known' good elevations to check
your DEM.

a.

DEM's are never accurate whether you produce from a georeferenced
map or from other sources .
Still a simple way can be that you take some precise GPS
observations at the undulating areas and refilter the DEM.