Marimo notebook to get ECMWF ENS data from Dynamical

[JackKelly]

TODO:

• Open Dynamical ECMWF ENS IceChunk
• Crop data for GB
• Plot temperature map (to check cropping is correct)
• Plot time series of all ensemble members
• Specify a list of the H3 indicies for GB (including a bit of off-shore), for a given H3 resolution (nearest to ECMWF ENS res)
• For each H3 hexagon, compute the weighting to compute the average NWP value across all the NWP grid boxes that lie under that hexagon.
• Use these weightings to compute the NWP per hexagon, and save into a Polars DataFrame. Columns:
  • init datetime
  • valid datetime
  • H3 index
  • ensemble member
  • temperature_2m_degrees_C
  • wind_speed_meters_per_sec
  • wind_direction_radians
  • etc.
• Find mins and maxes to scale each param to [0, 255]
• Convert to uint8
• Save as Delta Lake format using Polars. Can we partition by datetime without first computing year, month, etc?
• Save the mins and maxes as metadata in the Parquet files? (How does this work with Delta Lake?)
• Save the units as metadata, too?