Feature request: Bilinear or Tri-linear resampling for G2G

[songhan89]

Hi,

Had a look at the true colour and thought the native/nearerest neighbour sampling makes the edges of features a little bit hard. I add a slight hint of blur to my image, perhaps closer to what a bilinear or a tri-linear sampling might look like.

Just a matter of preference.

Would be good this feature can be added next.

[djhoese]

Thanks for the request. The bilinear interpolation is definitely possible and likely wouldn't be too hard to add. @pnuu with the pyresample project is currently optimizing the bilinear resampling. Once this is done there is no reason G2G couldn't use it in the future.

[graemely]

Here is a reference bilinear interp implementation in Fortran90:
https://gitlab.ssec.wisc.edu/geocat/geocat/blob/master/src/interpolation.f90

[djhoese]

We're currently discussing this in a meeting. Like I said Satpy currently has an implementation of bilinear resampling but if I remember correctly it is quite memory intensive. After talking with @graemely I think I'm going to find a case where bilinear resampling improves the resulting image, compare the execution time and memory usage with nearest neighbor resampling, and then add bilinear resampling as an option if it seems acceptable. Otherwise, we may wait for me to reimplement a different bilinear resampling algorithm that can perform better.

[pnuu]

We also have the experimental bilinear interpolation in Pyresample which uses gradient search. It is faster than the "normal" nearest neighbour resampling with cached LUTs. But as said, it is still experimental.

[kathys]

Not something that needs to be supported in Version 1.2. Push to 1.3.