Clouds obstruct the view from space, in particular in the visible spectral range, which often is of primary interest. This is particularly true in humid tropical regions. Very rarely do we get a satellite image over Colombia that is actually 100% cloud-free. If I take an image of the Saharan desert or the Australian outback, chances are much higher to get cloud-free images on a regular basis. Hence, given that we are interested in the tropics, we have to deal with this “nuisance” in the best possible way.
Clouds that are fairly thick are opaque and can be detected and masked in a semi-automatic way. If clouds are patchy, one can also use their shadows in the detection algorithm, though at tropical latitudes this works well only near local sunrise and sunset times. Thin clouds, in particular thin stratospheric clouds, can be difficult to detect and to mask. So far mostly semi-automatic algorithms are being used where a well trained person would help with picking the clouds and providing quality control. Artificial Intelligence software based on machine learning algorithms with multi-layer neural networks are in development in many research centers around the world, but to my knowledge few if any are ready for unsupervised production use.
We are presently using semi-automatic classic cloud detection algorithms that one of us has previously used for state-wide mapping in Australia adapting them for local conditions. However, we are also testing and developing machine learning AI solutions.
Obviously, where a cloud is being detected, no imagery information is available. Affected areas are turned into ‘blind’ spots.