| Re: terrain normalisation [message #20492] |
Fri, 30 June 2000 00:00 |
Marcel Droz
Messages: 3
Registered: June 2000
|
Junior Member |
|
|
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
Thanks a lot for your explanations! I actually see the problem...but as
our studies will cover the whole alps - means from France to Austria...
- its rather difficult to find a <b>homogenous</b> DEM of the whole region,
especially with limited finacial resources.
<br>Do you know any organisation, research institute or whatever, who deals/dealt
with this kind of DEM (about 250m resolution or even better)?
<p>Thanks in advance, Marcel.
<p>richard hilton wrote:
<blockquote TYPE=CITE>I would be very careful with calculating the aspect
of a 1km DEM when you
<br>are trying to use a 1km AVHRR image for sun angle/aspect/slope calculations.
<br>You need to use a DEM that is of much higher resolution than that.
You only
<br>have "mean" height values for the 1km squares and not any information
about
<br>the slope/aspect. You can infer a slope/aspect from the surrounding
pixels
<br>but this can result in increasing the errors in using this value instead
of
<br>assuming that the alps are flat!!!!! (the information from the surrounding
<br>pixels cannot give subpixel infomation about what is happening inside
the
<br>central pixel.)
<p>eg.
<br> 10 10
10
<br> 10 5
10
<br> 10 10
10
<p>A 3x3 grid. you would assume that the mean slope/aspect for the central
<br>square are both 0 but.....
<p> 10 &nbs p;
10 &nbs p; 10
<br> 10 10
5   ; 0 10
<br> 10 &nbs p;
10 &nbs p; 10
<p>in this case there is a linear slope to the right (the extra 10 and
0 are
<br>the subpixel heights in the inner square in the extreme left and right
of
<br>the square)
<p>but in this case the slope is reversed:
<p> 10 &nbs p;
10 &nbs p; 10
<br> 10 0   ;
5   ; 10 10
<br> 10 &nbs p;
10 &nbs p; 10
<p>this is obviously a simple case but I hope you can see the difference.
If
<br>you are trying to calculate the sun angle then assuming the area to
be flat
<br>is potentially going to be more accurate (even in a mountainous area!!!).
<p>I think that you really need to be using a far heigher resolution DEM.
and
<br>then rebin the slope/aspects up to the 1km required resolution.
<p>We are currently doing a lot of research into existing 1km (30 arc-second
to
<br>be precise) DEMs (and creating our own, called ACE (Altimeter Corrected
<br>Heights)) in particular GLOBE_v1, GTOPO30 and the 5 arc-minute JGP95E
and
<br>have shown errors in heights of up to 1500m over parts of the world.
I must
<br>however stress that I haven't looked at the swiss alps. I've had a
look at
<br>the area, and the source data in GTOPO30 is DTED data. This is the
best type
<br>of data in the DEM, but this uses a 3 arc second pixel to represent
the
<br>whole 30 arc-second region and is not a mean over the whole area. This
gives
<br>vertical errors (according to GLOBE_v1) of 18-120m (they used the same
data
<br>but the stats are far more acuarate!!) What affect this will have on
a
<br>calculation of the slope/aspect is difficult to assess but I assume
it would
<br>be fairly devastating!!
<p>General features (rivers etc. always appear in the correct place
<br>(horizontally but not vertically) but the topography is often supersampled
<br>100-200m contours which gives rise to mathematical features and NOT
true
<br>representations of the land surface. There should be accurately surveyed
<br>maps of the swiss alps available since it is a developed country and
the
<br>surveying has probable been done very well, but how much these cost
and who
<br>owns them I'm afraid I don't know. I imagine that even a 50m resolution
DEM
<br>for a relatively small area would be incrediably expensive.
<p>I hope this helps
<p>Let me know if you want anymore information.
<p>Richard</blockquote>
</html>
|
|
|
|
| Re: terrain normalisation [message #20516 is a reply to message #20492] |
Wed, 28 June 2000 00:00  |
richard hilton
Messages: 7
Registered: May 2000
|
Junior Member |
|
|
I would be very careful with calculating the aspect of a 1km DEM when you
are trying to use a 1km AVHRR image for sun angle/aspect/slope calculations.
You need to use a DEM that is of much higher resolution than that. You only
have "mean" height values for the 1km squares and not any information about
the slope/aspect. You can infer a slope/aspect from the surrounding pixels
but this can result in increasing the errors in using this value instead of
assuming that the alps are flat!!!!! (the information from the surrounding
pixels cannot give subpixel infomation about what is happening inside the
central pixel.)
eg.
10 10 10
10 5 10
10 10 10
A 3x3 grid. you would assume that the mean slope/aspect for the central
square are both 0 but.....
10 10 10
10 10 5 0 10
10 10 10
in this case there is a linear slope to the right (the extra 10 and 0 are
the subpixel heights in the inner square in the extreme left and right of
the square)
but in this case the slope is reversed:
10 10 10
10 0 5 10 10
10 10 10
this is obviously a simple case but I hope you can see the difference. If
you are trying to calculate the sun angle then assuming the area to be flat
is potentially going to be more accurate (even in a mountainous area!!!).
I think that you really need to be using a far heigher resolution DEM. and
then rebin the slope/aspects up to the 1km required resolution.
We are currently doing a lot of research into existing 1km (30 arc-second to
be precise) DEMs (and creating our own, called ACE (Altimeter Corrected
Heights)) in particular GLOBE_v1, GTOPO30 and the 5 arc-minute JGP95E and
have shown errors in heights of up to 1500m over parts of the world. I must
however stress that I haven't looked at the swiss alps. I've had a look at
the area, and the source data in GTOPO30 is DTED data. This is the best type
of data in the DEM, but this uses a 3 arc second pixel to represent the
whole 30 arc-second region and is not a mean over the whole area. This gives
vertical errors (according to GLOBE_v1) of 18-120m (they used the same data
but the stats are far more acuarate!!) What affect this will have on a
calculation of the slope/aspect is difficult to assess but I assume it would
be fairly devastating!!
General features (rivers etc. always appear in the correct place
(horizontally but not vertically) but the topography is often supersampled
100-200m contours which gives rise to mathematical features and NOT true
representations of the land surface. There should be accurately surveyed
maps of the swiss alps available since it is a developed country and the
surveying has probable been done very well, but how much these cost and who
owns them I'm afraid I don't know. I imagine that even a 50m resolution DEM
for a relatively small area would be incrediably expensive.
I hope this helps
Let me know if you want anymore information.
Richard
|
|
|
|
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
