[IDEA] Compositing ir_sanwdich with geo_color

[guidocioni]

I always dreamed of being able to combine geo_color with ir_sandwich because the former becomes kind of saturated in areas with many reflective cloud tops (deep convection) while the latter doesn't really capture the contrast between land and stratus clouds. After giving claude some food for thoughts (and my code as example) it was able to create a result that, to my eye, does not look too bad...

Here is an example for some strong thunderstorms that formed some days ago over Senegal.

Original geo_color

Original ir_sanwdich

Merged product

I (we) had to tweak something to get the right overlay effect, but it was more straightforward than I thought

The pseudo code looks something like this

      geo_color_image = get_enhanced_image(fci_scene_r["geo_color"]).data.compute()
      sandwich_image = (
          get_enhanced_image(fci_scene_r["ir_sandwich_with_night_colorized_ir_clouds"])
          .data.compute()
          .transpose("y", "x", "bands")
      )
      bt = fci_scene_r["ir_105"].compute().values  # Kelvin, native resolution/grid
      valid = ~np.isnan(bt)

      # NaN/off-swath BT is forced fully transparent rather than left to propagate
      # into the alpha channel -- that produced a stray blue-ish ring at the mask
      # edge in an earlier version (NaN alpha smeared by interpolation).
      alpha = np.where(
          valid, np.clip((IR_WARM_K - bt) / (IR_WARM_K - IR_COLD_K), 0.0, 1.0), 0.0
      )
      ir_rgba = np.concatenate(
          [sandwich_image.values, np.zeros_like(sandwich_image.values[..., :1])], axis=-1
      )
      ir_rgba[..., 3] = alpha

      geo_color_image.plot.imshow(
          vmin=0, vmax=1, add_colorbar=False, rgb="bands", ax=ax,
          transform=area_def.to_cartopy_crs(), zorder=4, interpolation="spline36",
      )
      ax.imshow(
          ir_rgba,
          transform=area_def.to_cartopy_crs(),
          extent=[x_min, x_max, y_min, y_max],
          origin="upper",
          zorder=6,
          # nearest, not spline36: cubic/spline kernels ring at the sharp alpha edge
          # on a non-monotonic (diverging) colormap like Spectral, producing false
          # hues instead of a smooth fade.
          interpolation="nearest",
      )

I believe this composite could not be added as-is to the YAML, but do you think is of any interest? Or did I duplicate something that already exists? :)

Note: there are still some graphical artifacts at cloud edges due to the combination of both FDHSI and HRFI. Haven't found a solution for that yet.

[gerritholl]

This looks like a nice addition! Your examples are daytime, where geo color is identical to true color. Does it also affect the night part, where geo color is entirely different, and IR sandwich is just a colorisation of the IR window channel? If it's day time only I would call it "a composition of IR sandwich with true color", not geo color.

I need to think if this can be done with existing compositors. My first thought is "no", but sometimes compositors can do more than we think at first.

[guidocioni]

I had the same doubt but it seems to be fine at night because sandwich only overlays really high clouds. Here is a full animation

geo_color_ir_overlay.mp4

Still I think it should be tested on a wider area.

[strandgren]

Actually I was working on improving the ir_sandwich composite last week since it's currently very dark around the cold cloud tops, with the result that we rarely use it ourselves which I think is a pity. My solution was to use the masking and background compositors which allows you to overlay the colourized part of the ir_sandwich on top of any other composite, including geo_color For my testing I just overlaid it on top of the VIS0.6 data to stay close to the original ir_sandwich composite. You can see an example here: https://sftp.eumetsat.int/public/file/m4crpuhrd0qqi7aqdiomyw/FCI_ir_sandwich_with_colorized_night_layer-FR-12.mp4

This could be achieved with yaml/configuration changes only.

I was planning to create an issue about this when I'm back from holidays in about two weeks time. It also includes an extended colormap in order to have more intense colours both in the tropics and mid-latitudes. The current one (Spectral) typically looks quite lame for European storms.

[guidocioni]

Actually I was working on improving the ir_sandwich composite last week since it's currently very dark around the cold cloud tops, with the result that we rarely use it ourselves which I think is a pity. My solution was to use the masking and background compositors which allows you to overlay the colourized part of the ir_sandwich on top of any other composite, including geo_color For my testing I just overlaid it on top of the VIS0.6 data to stay close to the original ir_sandwich composite. You can see an example here: https://sftp.eumetsat.int/public/file/m4crpuhrd0qqi7aqdiomyw/FCI_ir_sandwich_with_colorized_night_layer-FR-12.mp4

This could be achieved with yaml/configuration changes only.

I was planning to create an issue about this when I'm back from holidays in about two weeks time. It also includes an extended colormap in order to have more intense colours both in the tropics and mid-latitudes. The current one (Spectral) typically looks quite lame for European storms.

Looks really promising, and this is exactly what I was looking for. Claude seemed to implied the logic about the masking could not be implemented cleanly in the YAML alone....:) If you have time could you paste the modification you had to made?

[strandgren]

Actually I was working on improving the ir_sandwich composite last week since it's currently very dark around the cold cloud tops, with the result that we rarely use it ourselves which I think is a pity. My solution was to use the masking and background compositors which allows you to overlay the colourized part of the ir_sandwich on top of any other composite, including geo_color For my testing I just overlaid it on top of the VIS0.6 data to stay close to the original ir_sandwich composite. You can see an example here: https://sftp.eumetsat.int/public/file/m4crpuhrd0qqi7aqdiomyw/FCI_ir_sandwich_with_colorized_night_layer-FR-12.mp4
This could be achieved with yaml/configuration changes only.
I was planning to create an issue about this when I'm back from holidays in about two weeks time. It also includes an extended colormap in order to have more intense colours both in the tropics and mid-latitudes. The current one (Spectral) typically looks quite lame for European storms.

If you have time could you paste the modification you had to made?

Something like this should work for the daytime part:

  ir_sandwich:
    compositor: !!python/name:satpy.composites.SandwichCompositor
    standard_name: ir_sandwich
    prerequisites:
      - name: vis_06
        modifiers: [effective_solar_pathlength_corrected]
      - name: colorized_ir_clouds

  ir_sandwich_masked:
    compositor: !!python/name:satpy.composites.MaskingCompositor
    standard_name: image_ready
    conditions:
      - method: greater_equal
        value: 253.15
        transparency: 100
      - method: isnan
        transparency: 100
    prerequisites:
      - name: ir_sandwich  # Composite
      - name: ir_105  # Data used in masking

  true_color_sandwich:
    compositor: !!python/name:satpy.composites.BackgroundCompositor
    standard_name: image_ready
    prerequisites:
      - ir_sandwich_masked
      - true_color

For the geo_color, including night-time layer, you also need to mask the colorized_ir_clouds composite and merge the two colorized daytime/nighttime layers together, before putting the combined layer on-top of geo_color. Will create the issue with full context when I'm back, but hopefully this can be helpful in the meantime 🙂

[alexhalbig]

Hi all, I just came across your discussion and I really like the idea of the combined IR-Sandwich on top of True/Geo Color RGB.

"The current one (Spectral) typically looks quite lame for European storms."

I agree, thats why I already tried to improve the "spectral" color scheme for cases with very cold cloud tops over Europe and around. That's what I got. More reddish colours for cold structures. I notized a drawback in my current enhancement, as "mid-temperated" clouds are therefore more saturated/yellowish.

Thats my enhancement, in case you are interested:

  colorized_ir_clouds_eu:
    standard_name: colorized_ir_clouds_eu
    operations:
      - name: colorize
        method: !!python/name:satpy.enhancements.colormap.colorize
        kwargs:
          palettes:
            - {colors: spectral, min_value: 200.15, max_value: 253.149999}
            - {colors: greys, min_value: 253.15, max_value: 303.15}

[strandgren]

I just opened issue #3440 about some possible improvements of the ir_sandwich imagery and downstream us of it - happy to get your feedback there 🙂