[GRASS5] Re: [GRASSLIST:7895] Re: Hutchinson's Adaptive Alogrithm for sound DEMs?

[Jaro Hofierka]

Maciek Sieczka wrote:

>> The recent work by  Tomas Cebecauer and others (See "Processing digital
>> terrain models by  regularized spline with tension: tuning interpolation
>> parameters for  different input datasets" from the proceedings to the 
>> 2002
>> GRASS  conference.) shows how v.surf.rst can be used in a method similar
>> to  ANUDEM to enforce proper drainage networks, by adding a "terrain
>> skeleton".
> 
> 
> I'm affraid the method described in Cebecauer's paper is *completely*
> diffferent from the one used in ANUDEM. ANUDEM supports watercourse *lines*
> and elevation fault *lines*. While for v.surf.rst you have to digitise them
> as *points*. What's really bad, each such point has to be labelled for
> elevation before you feed it into v.surf.rst. So actually you have to do 
> the
> interpolation of fault lines, ridges and watercourses manually, before 
> you can make v.surf.rst use this crucial information. I find it strange, 
> because a DEM interpolation program should be able to do it alone. At 
> least that's the way I see it.
> 
> Summarrising, including watercourses and fault lines in ANUDEM (and 
> ridges in CatchmenetSIM) is quick and easy. In v.surf.rst extremely time 
> consuming, thus virtually impossible, i.e. a non-existant feature.
> 
> Example:
> 
> There is stream 10000 meters long.
> How long would it take one to digitise it as a line and include in ANUDEM?
> Answer: a minute :) ?
> 
> And how long is it to digitise points dense enough and label each for
> elevation along the 10 km long watercourse for v.surf.rst? Ok, one could
> digitise the line and convert into points then, half work less work. But
> still, *each* point has to be labelled for elevation now. A real problem -
> interpolate manually the elavation of all the points and label each - e.g.
> 1000 points in case of 10 m resolution, 10000 points if res=1 m.
> 
> Answer:  say, 5 sesonds a point. In case of 10 m resolution this gives us 83
> minutes. If one wants the 1 m resolution, he'd have to settle for something
> about 14 *hours*. Who dares?

Hi,

The method described in Cebecauer et al. 2002 has 2 parts. So called 
"terrain skeleton" consists of mostly vector lines digitized from maps 
(ridges, valley lines, etc.). The vertices for these lines are 
automatically "densified" to a regular step, i.e. linearly interpolated 
along the line, using a script (currently available only in ArcView GIS).
If I remember correctly, the script takes elevation values from contours 
and interpolates values for points automatically generated between the 
contours.
So, there is no need for "massive" manual editing for these points.
At that time we were using just "site" version of the RST method, so we 
needed points to enforce the shape of resulting surface.

Another reason for this was that usually we interpolate such data with 
variable smoothing option, i.e. input dataset consists of several parts, 
each assigned with different smoothing parameters. So, for example, 
"skeleton lines" have different smoothing than contours.

Jaro