[GRASSLIST:3404] Re: srtm: correct big holes with r.fillnulls

[Markus Neteler]

On Tue, May 11, 2004 at 03:46:01PM -2300, tatel at euskalnet.net wrote:
> Hi Markus and all subscribers,
> 
> 
> Markus wrote:
> >PS: It seems that you have to reproject to non-LatLong before using r.fillnulls.

A few days ago the LatLong bug in RST has been solved.
It was "just" an integer variable which should have been double.
Fixed for 5.3 and 5.7.

> 
> r.fillnulls worked fine for me *in* a Lat-Lon location.

It worked sometimes, but not always - depending on the hole shape
and diameter.

> In fact, after 
> reprojecting to a not Lat-Lon location (e.g, UTM) process takes much more 
> time because there are now null borders too. (I was correcting an entire 
> tile)

Since the problem is solved, please try the current GRASS.

> Filled a very big hole (many Kms) and a lot of medium-little holes in 
> N43W005 srtm3 tile corresponding to Picos de Europa mountains.
> 
> I found imported srtm data have NULLs set to zero so I needed to do first 
> (to get r.fillnulls working as expected)
> 
> r.mapcalc srtm3=if'(srtm3==0,null(),srtm3)'

Strange. I'll look into this.

> 
> It is easy to get better results by merging srtm30 data at correct 
> resolution.

In case srtm30 data are available .... (only US AFAIK).

> g.region res=00:00:30 
> r.to.sites -a input=srtm30 output=srtm30.sites
> g.region res=00:00:03
> s.to.rast input=srtm30.sites output=srtm30cells
> r.mapcalc new=if'(isnull(srtm3),srtm30cells,srtm3)'
> 
> So we get a "grid" of isolated cells each 900m into the NULL zone(s). Now 
> we can run 
> 
> r.fillnulls input=new output=corrected 
> 
> In this case, without srtm30 data there's a valley runing S-N which 
> becomes divided in a little valley and a closed depresion. Using the 
> merged srtm30 data the entire valley is correct.
> 
> s.surf.rst as invoked in r.fillnulls worked very well in (maybe not so) 
> little holes but produced ugly results (segmentation) with only a 
> non-NULL cell each 900 m  
> 
> I canibalized r.fillnulls and found that invoking s.surf.rst with npmin=
> 1000 produced some warnings but worked quite fast with much more smoothed 
> segments.

I think I still have to upload a change to r.fillnulls to follow
the latest changes in s.surf.rst (v.surf.rst in 5.7).

> Unfortunately hard disk crashed next day so still I haven't modified 
> r.fillnulls. Im writing this on my old powerbook:-((
> 
> Hope this helps somebody (probably as newbie as me), and thanks to all 
> who helped me in correcting srtm data.
> 
> Roman

Markus