Support for multiscale data (post extraction)

[leoglonz]

Summary

High-performance, CONUS-scale models currently depend on multiscale data, which is processed differently than single-scale data. Initial steps have been taken to adapt an example of this (GAGES II + MERIT), but there is still room for simplification, generalization, and documentation of methods.

Actions

• Preprocessing for Trainers needs to be added to hydroDL2.
  • See current PMI implementation hacked together here.
  • Add documentation for methods and rationales in the preprocessing.
• Clean up code with better nomenclature and potentially improvements for efficiency.

Considerations

• Data processing for multiscale exists, but needs refinement.
• Methods and rationales need to be documented.
• Currently only exists for GAGES II + MERIT. Can the methods generalizable to other datasets? Probably not, but it's worth identifying boilerplate.

Assignments: @Yalan-Song, @taddyb, @hiyuji98, @leoglonz