How to export clouds from satellite imagery?

Question

Image processing

From Worldview you can get daily surface shots. You need to get a layer with clouds, without taking into account the surface of the Earth.
The algorithm is as follows:
- get tiles without clouds
- subtract the one without clouds from a certain tile with clouds
- the resulting tile will contain pixels whose brightness is equal to the difference on tiles with and without clouds
The following problems appear:
- in some places there is snow cover, on clouds cannot be distinguished in the background
- there are joints in the images in which the clouds do not match
- reflection of the sun's rays on the water surface, which can be taken as clouds
Example: 8

Taking images from several days in a certain tile, I built a new image consisting of only the minimum brightness from each corresponding pixel of the images. Thus, it is possible to get a tile without clouds, but there is also noise caused by the presence of clouds on each of the images (in other words, it was cloudy at a certain coordinate for all the days for which the images were taken, even for 100 days)
No clouds, but with noise : **

If we subtract the second** from the first* image, we get:

It seems to be not bad, but the loss of brightness of the clouds is visible against the background of the land, so I tried to calculate the difference using a slightly different formula in order to maintain high brightness in the original image with clouds:

As a result, the problem remains with the reflection on the surface of the water, as well as the snow cover, due to which the clouds are absorbed when subtracting

Answer 1

You definitely need data from different parts of the spectrum, then you will see both clouds and snow and be able to somehow separate them.