Displaying continuous data with multiple transparency values?

[gadomski]

Tl:dr; Is there a way to define a continuous colormap while still mapping multiple values to transparency?

We're using rio-cogeo (via titiler) to display COG-ified MODIS products. Some MODIS products, e.g. MOD17A3HGF, have continuous data that we'd like to display with a continuous colormap. However, these products also have multiple "fill values" that make displaying "nodata" as transparent pixels tricky. E.g. MOD17A3HGF has the following two continuous variables:

Data Sets	Meaning	Units	Date	Type	Valid Range	Scale Factor
Gpp_500m	8-day total GPP	kgC/m2	/8d	int16	0 ~ 30000	0.0001
PsnNet_500m	8-day total Net Photosynthesis	kgC/m2	/8d	int16	-30000 ~ 30000	0.0001

These int16 values also have "fill values" at the top end of their domain -- we want to display all of these values as transparent pixels:

32767 = _Fillvalue
32766 = land cover assigned as perennial salt or Water bodies
32765 = land cover assigned as barren,sparse veg (rock,tundra,desert)
32764 = land cover assigned as perennial snow,ice.
32763 = land cover assigned as "permanent" wetlands/inundated marshland
32762 = land cover assigned as urban/built-up
32761 = land cover assigned as "unclassified" or (not able to determine)

For now, we're using an intervals colormap to get close to what we want, e.g.:

[
    [
        [ 0, 3750 ], [ 247, 252, 253, 255 ]
    ],
    [
        [ 3750, 7500 ], [ 229, 245, 249, 255 ]
    ],
    [
        [ 7500, 11250 ], [ 204, 236, 230, 255 ]
    ],
    [
        [ 11250, 15000 ], [ 153, 216, 201, 255 ]
    ],
    [
        [ 15000, 18750 ], [ 102, 194, 164, 255 ]
    ],
    [
        [ 18750, 22500 ], [ 65, 174, 118, 255 ]
    ],
    [
        [ 22500, 26250 ], [ 35, 139, 69, 255 ]
    ],
    [
        [ 26250, 30000 ], [ 0, 88, 36, 255 ]
    ],
    [
        [ 30000, 32766 ], [ 0, 0, 0, 0 ]
    ]
]

Is there a way to define a continuous colormap while still mapping multiple values to transparency? If not, is this something that could be implemented? I'm happy to dig in an give it a shot if its something that's possible.

[vincentsarago]

👋 @gadomski do you mean that you want:

• to hard code a specific colormap within the data?

I'm not sure about this but maybe you could try to create a discrete colormap (mapping every values of the Uint16 data to a RGBA values), and use [0, 0, 0, 0] for the values you want to mask.

• to dynamically colourize the data using an external colormap?

This is more a rio-tiler/titiler question and I don't see what is wrong with what you are using right now.

Worth noting that you are using a scale factor which if use will transform the data before applying the colormap (so the colormap values won't match).

[gadomski]

you could try to create a discrete colormap

The issue w/ this is that there are 30000 discrete values that need to be colorized, so a continuous map feels much more appropriate.

to dynamically colourize the data using an external colormap?

This is our goal, but if I understand correctly, the colormaps are being passed down to rio-cogeo under the hood, so any new colormap features would need to be implemented at the rio-cogeo level?

I don't see what is wrong with what you are using right now.

The issue with what we're doing now is that we're forced to visualize our data in (this case) eight color bands, instead of a continuous color ramp. It's not a deal-breaker but I'd be interested in exploring a "continuous colormap with multiple transparent values" solution if it's technically possible

Worth noting that you are using a scale factor which if use will transform the data before applying the colormap (so the colormap values won't match).

For our purposes titiler doesn't apply the scaling unless we ask it to via the unscale parameter, so for now we're specifying the colormap in terms of the raw int16 values.

[vincentsarago]

to dynamically colourize the data using an external colormap?

This is our goal, but if I understand correctly, the colormaps are being passed down to rio-cogeo under the hood, so any new colormap features would need to be implemented at the rio-cogeo level?

There are 2 things here, either the COG has an internal colormap (https://github.com/cogeotiff/rio-tiler/blob/master/rio_tiler/io/cogeo.py#L195-L201, https://github.com/developmentseed/titiler/blob/master/src/titiler/core/titiler/core/factory.py#L495) which can be added with GDAL or rio-cogeo Or you dynamically tell titiler/rio-tiler to colourize the data with a given colormap as defined by rio-tiler (https://cogeotiff.github.io/rio-tiler/colormap/) with the possibility to either be:

• discrete (with custom entries, not limited to uint8 type) - NOT SUPPORTED BY GDAL
• linear (with 256 values from 0 to 255)
• Intervals colormaps (in form of [[[Min, Max], [R, G, B, A]], ...] - NOT SUPPORTED BY GDAL

The issue with what we're doing now is that we're forced to visualize our data in (this case) eight color bands, instead of a continuous color ramp. It's not a deal-breaker but I'd be interested in exploring a "continuous colormap with multiple transparent values" solution if it's technically possible

Well you just need to create more intervals 🤷‍♂️

note: NOT SUPPORTED BY GDAL means that the format of the colormap is not supported by GDAL and thus can't be hardcoded in the COG directly.

[gadomski]

note: NOT SUPPORTED BY GDAL means that the format of the colormap is not supported by GDAL and thus can't be hardcoded in the COG directly.

Yup, we're definitely interested in a dynamic solution -- modifying the COGs to add a colormap isn't really within scope for what we're doing as far as I understand.

Well you just need to create more intervals 🤷‍♂️

Fair enough 👍🏽. Thanks!