Re: Offset threshold problem

> > For what I've seen so far the STRM data incorporates terrain
> > elevation due to stables, farm houses, trees (or rather small
> > woods), tall buildings, etc.
> >
> > So in reality the data is too accurate for our needs.
> 
> 
> Yes, when you start digging into this stuff it get's complicated
> quick.  What we *want* is the elevation of the ground for the entire
> earth.  What we get is radar reflections off of what's ever there ...
> often we can see evidence of objects in the data.  Are we getting the
> tops of trees or the bottom?  The further you dig into this stuff, the
> harder it get's.  With some sensors you might get a primary reflection
> off the top of trees, a secondary reflection of the lower level bushes
> and once in a while a 3rd refelection off the ground.  Now try to sort
> that out. :-)


Make a grid, maybe of 3x3 3arcsec cells (~275m a side). The elevation
value used for the DEM is the lowest data point within each 3x3 cell.
You lose detail, but you get a much cleaner end product. Life's a
compromise...

Either that or run a low pass filter on the data to remove all the
spikes. If above Max change of slope abs(d^2z/dx^2) threshold, use an
average of surrounding cells; that sort of thing.

Many gov't forestry types are working on this, but it's still new. Keep
an eye out for brand new journal articles .. it's a common problem, esp.
with more & more LIDAR data arriving than people know what to do with.


good luck,
Hamish