[GRASS-user] about resampling DEMs and DEM-derived parameters

[Daniel Victoria]

Not sure for slope and asect but for flow direction and accumulation,
you run into trouble if you resample the dem and then run the
algorithms for flowdir. There are actually some interesting works
around saying that the best thing to do in this case would be to get
the flowdir and accumulation on a finer grid and then use a special
method to rescale that.

This paper comes to mind

Olivera, F., M. S. Lear, J. S. Famiglietti, and K. Asante. “Extracting
Low-resolution River Networks from High-resolution Digital Elevation
Models.” Water Resources Research 38, no. 11 (2002): 1–8.

Cheers
Daniel

On Mon, Jun 4, 2012 at 2:44 PM, Glynn Clements <glynn at gclements.plus.com> wrote:
>
> Carlos Grohmann wrote:
>
>> The other day something popped on my mind, which I think could end
>> in some research paper, but before I venture into it I decided to
>> ask you guys about it. I did some googling and got nothing.
>>
>> So, considering a large area DEM: if I want its derivatives (slope,
>> aspect, etc) in large-scale (larger than the original DEM), which
>> one is best, to resample the DEM and then calculate the derivatives,
>> or to calculate the derivatives and resample them? Much difference
>> expected? The same?
>
> You get better results if you know what you're actually trying to
> measure. The concept of a derivative is only meaningful for a
> continuous surface, which a DEM isn't. Do you want the actual
> derivative at a specific point (which may be the side of a rock), or
> the slope of some "best-fit" plane?
>
> It also helps if you know what the DEM represents. E.g. is the
> elevation a sample at a specific point within the cell, or an average
> over the area of the cell?
>
> As a general guide, you probably want to apply some form of low-pass
> filter to the data, so that the derivatives at each point are
> calculated for a local best-fit curve.
>
> 7.0 has r.resamp.filter, which performs resampling with an analytic
> (continuous) kernel. In 6.x, you can obtain similar results using a
> combination of upsampling, filtering with r.neighbors or r.mfilter.fp,
> then downsampling.
>
> --
> Glynn Clements <glynn at gclements.plus.com>
> _______________________________________________
> grass-user mailing list
> grass-user at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/grass-user