Building the training data¶
HealDA trains on two datasets, each built by an ETL in scripts/etl/:
| dataset | HealDA reads it as | location setting | ETL | steps |
|---|---|---|---|---|
| ERA5 0.25° hourly | targets (Era5HourlyH5) |
ERA5_HOURLY_73CH (or ERA5_HOURLY_ROOT), ERA5_HOURLY_EXT_ROOT |
scripts/etl/era5/ |
Building the ERA5 target files |
| NNJA observations | model inputs (the NNJA observation loaders) | NNJA_ROOT |
scripts/etl/nnja/ |
Building the NNJA archive |