I have three questions regarding this topic:

(1) What should I consider as inputs (for a machine learning model) when attempting crop type classification. Especially, when it comes to features, what strategy should I follow to attempt such a task.

(2) I noticed that some people use image composites and NDVI composites, but they don’t provide the reasons for choosing such techniques, and to be honest this terminology is still not clear to me.

(3) When using Sentinel L2A data, what kind of preprocessing techniques should I consider before using the data?