Detailed Description

Currently we don't have any satellite normalisation, as its between 0 and 1, but we should have, like we did in ocf-datapipes. This is because some channels has very similar values, and normalising by mean and std, means the ML model can learn more easily the behaviour of that channel

Context

• old scaling is done with some constant values here
• before we had it as an option, but we should do this as default
• Similar to NWP normalization

Possible Implementation

• add satellite normalization in torch_datasets here