[GRASS-dev] r.fill.gaps porting

[Benjamin Ducke]

On 16/04/17 03:40, Vaclav Petras wrote:
> 
> On Sat, Apr 15, 2017 at 8:32 AM, Benjamin Ducke <benducke at fastmail.fm
> <mailto:benducke at fastmail.fm>> wrote:
> 
>     The thing is: I originally developed this module for
>     gradiometer data. That data is very noisy and has
>     high local variation. Interpolating that type of
>     data while preserving original measurements will
>     usually result in unsatisfactory output. Also note
>     that the smoothing increases with the interpolation
>     radius. For tiny holes in LiDAR data, try "distance=1"
>     or "2". Of course, LiDAR data probably has different
>     properties and does not need low-pass filtering
>     like gradiometer data. So try "-p" to preserve the
>     original values.
> 
>     An alternative would be to invert the behaviour of
>     the module and assume "-p" by default.
> 
>     I don't work with LiDAR data myself, but I would very
>     interested to know your results!
> 
> 
> 
> I've added a figure which shows the r.slope.aspect products from the
> surface without and with -p. You can see that the smoothing gives
> significantly different results and probably better ones. I had the same
> experience in past when trying to patch result of r.neighbors with the
> original raster. Perhaps using smaller distance as you say or greater
> power would help, but I haven't tested that yet.

I find the results for "curvature" most interesting.
In the unsmoothed LiDAR data, that metric is basically useless.

If I understand correctly, then slope, aspect
and curvature are all computed within a small
neighbourhood of cells. So if LiDAR data is
also noisy by nature, then it is not surprising
that these metrics are all distorted by strongly
fluctuating local means.

Please try experimenting with the power parameter,
if you have the time. In theory, this should allow
you to interpolate over larger distances with less
of a low-pass effect (or the other way around,
depending on what you want to achieve).

> 
> I managed to commit just the figure its caption. If somebody gets to it,
> please extent the section.

I will update the HTML page, including a more
prominent mentioning of the default smoothing.
I also found some small typos that need fixing.

Best,

Ben