[Qgis-user] nodata and raster calc

[qgis-user at stripfamily.net]

On 2/10/2019 7:11 PM, Nyall Dawson wrote:
> Hey all,
>
> I'm seeking some feedback on whether our current handling of nodata
> pixels in the raster calculator is correct, or (my suspicion) annoying
> and limiting.
>
> Currently, we treat nodata pixels the same as a "NULL" value in SQL --
> so ANY function or operation with nodata as an input results in a
> nodata output.
>
> This means:
>
> - nodata + 1 = nodata
> - nodata * 1 = nodata
> - nodata AND 1 = nodata
> - nodata OR 1 = nodata

As much as I understand the desire to manipulate nodata pixels, I'm not
sure what I would expect for the first two cases you list. I would hope
the result would be the same regardless of the internal representation
of nodata, so that rules out just adding 1 to the value.  My natural
inclination I would be to expect the AND case to return nodata, esp if
the OR case is supposed to return 1, but others may have different
expectations/intuition.

I would encourage the use of explicit conditionals to deal with nodata
cases and extend the raster calculator to accept "nodata" as a literal 
in conditionals so that the user doesn't have to know what the nodata
value actually is.
For example if you want nodata + 1 to be  1 in your first example above

      (raster at 1 == 'nodata') * 1 + (raster at 1 != 'nodata') * (raster at 1 + 1)

For what I would expect to be the rare case where you want to manipulate
the actual value of nodata, I suppose we'll need either a function that
returns the nodata value  - eval(nodata) or another literal, eg,
nodata_value.
I find that idea of operating on the nodata value itself to be rather
dangerous as your result changes with a change in the representation of
nodata, which seems semantically meaningless.

David Strip