I just received a strong letter from the developer of the algorithm
in question. Please read the follows exchange before proceding
to examine my results.
From hgreen Tue Mar 23 17:22:49 1999
To: ecoregio@t-online.de, hutch@cres20.anu.edu.au
Subject: Re: topogrid/anudem
Cc: apatel@esri.com, skopp@esri.com
> From hutch@cres20.anu.edu.au Tue Mar 23 16:30 PST 1999
> From: Mike Hutchinson
> Date: Wed, 24 Mar 1999 11:33:27 +1100 (EST)
> To: ecoregio
> cc: apatel@esri.com, Steve Kopp ,
> Harvey Greenberg
> Subject: topogrid/anudem
>
> Dear Olaf
>
> I have seen that article!
>
> http://duff.geology.washington.edu/topotest/index.html
>
> The "data" are so unrealistic that I have thought it not worthy of a reply
> to be honest. I can see that I will need to do more.
>
> I would have appreciated a direct contact from the author(s) to seek my
> response, or at least the response of an impartial referee, before
> releasing their conclusions to the world!
I apologize for not having contacted you earlier. I thought that I was
discussing an acknowledged feature of topogrid.
>
> For now, note that each patch of totally straight data contours were
> simply pieced together and gridded as as whole - OF COURSE TOPOGRID/ANUDEM
> tried to splice together this mish mash. TOPOGRID/ANUDEM makes use of ALL
> the data - it does not work on each bit in isolation.
I thought that the distance between patches would isolate them sufficiently.
I will check these results more carefully.
>
> If the data pieces had been processed separately, the result would have
> been separate planar grids, honouring the data quite well - if the usual
> current default smoothing parameters for TOPOGRID/ANUDEM were chosen.
> There would not even have been any appreciable biasing towards contour
> elevations!
I undertook my study of patches because previously attempts to build DEMs
from mathematically defined contours, working one hillslope at a time,
had been unsatisfactory. I will present details as soon as I have an
opportunity.
>
> REAL contours are curvilinear - the small bias that many are worried about
> arises with curvilinear contours. I repeat that most if not all
> applications of DEMs do not actually USE the elevation histogram, and that
> plan curvature and slopes are well represented, especially if the DEM
> resolution is chosen appropriately. The bias in profile curvature is
> acknowledged (but lessened by appropriate choice of DEM resolution) - and
> is being addressed (subject to the serve limits on my time - academia is
> not the free and easy occupation it once was).
My concern is not so much for the elevation histogram as for local hillslope.
A small patch of steep hillslope can precipitate a landslide, and the way
such physical features are represented in a DEM is our great importance to
many people. (And wetness is also of great importance, for which the work
of ANU must be acknowledged.)
>
> The relatively good representation of terrain shape by ANUDEM has in fact
> encouraged more refined assessments of DEM shape. Many early elevation
> interpolators in fact had worse elevation histograms, and much worse plan
> curvature (which defines surface drainage), unless the interpolated DEMs
> were individually and semi-subjectively massaged - the former practice of
> some agencies as I understand it.
If I had another interpolator that I preferred, I would have mentioned it.
>
> I would be interested to see any documented comparisons between any
> currently supported techniques applied to REALISTIC contour, point and
> streamline data. Results are always as much dependent on the data as on
> the processing technique! Does anyone remember "GIGO"?
I will pursue this matter further when I have time. Until then, I will, if
you do not object, post this correspondence in front of my web page.
>
> Feedback very welcome.
>
>
> Sincerely,
>
> Dr Michael Hutchinson
> Centre for Resource and Environmental Studies
> Australian National University
> Canberra ACT 0200
> Australia
> Email: hutch@cres.anu.edu.au
> Fax: +61-2-6249-0757
>
>
Harvey Greeenberg hgreen@u.washington.edu 162 Johnson
206-685-7981 http://duff.geology.washington.edu/
University of Washington
Geological Sciences
Box 351310
Seattle WA 98195-1310
Main Mountain Drainage
Research Group Page