Notes for upgrading to the 0p25_1hr model

[darksidelemm]

Tawhiri currently uses the 0.5 degree spatial resolution, 3 hour time resolution model (0p50_3hr).

A good step forward in prediction accuracy (in some circumstances at least...) would be to move to the 0.25 degree spatial resolution, 1 hour time resolution GFS model (0p25_1hr). This will require changes in both the downloader and this repository.

I have made notes on the likely required changes in the downloader here.

The overall changes are are:

• General increase in dataset size. We go from ~9GB to ~80GB per dataset file. The dataset is a multi-dimensional array of float32s, with axes:
  • hour - Time into the model, in hours
  • level - Pressure level
  • variables - what variables we have for each position (we use UGRD, VGRD and HGT)
  • Latitude
  • Longitude
• Changing from a time-step that steps by 3 hours across the entire model time range, to one that starts out with 1 hour steps, then changes to 3 hour steps at +120 hours.

Some notes on where changes will need to be made in the code:

• The dimensions of the dataset are defined here. This will change to 145, 47, 3, 721, 1440.
• If there are any extra pressure levels available to us, they would need to be added here
• The 'lookups' for each axis are defined here and will need to be adjusted for the new dataset sizes and values (e.g. for hour this will be something like: [x for x in range(0, 121)] + [x for x in range(123, 192+3, 3)]

In the interpolator code, a few changes need to be made:

• Dataset dimensions here
• A bunch of hardcoded dimensions in pick3 here. The pick call for the hour will get a bit more complicated, and may need a separate function due to the change in step size partway through the times.
• Possible changes here - I think this refers to the number of pressure levels.