The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
ERA5 → CONUS404 Downscaling Data
Raw input (ERA5) and target (CONUS404) NetCDF files for the "Latent CorrDiff"
downscaling pipeline. Pairs with the model checkpoints in
mudhil/era5-conus404-diffusion-downscaling.
Contents
| Folder | Files | Size | Description |
|---|---|---|---|
era5_data/ |
era5_{YYYY}.nc, 41 files (1980–2020) |
~6 GB total | ERA5 input, CONUS subset, 0.25°, daily |
conus404_data/ |
conus404_yearly_{YYYY}.nc, 41 files (1980–2020) |
~2.4 TB total | CONUS404 target, 4km, daily |
ERA5 (era5_data/)
Rebuilt from the public ARCO-ERA5 mirror. Each file holds t2m, d2m,
u10, v10, sp, tp, and lsm (land-sea mask) at daily resolution over
a CONUS subset. Instantaneous variables (t2m, d2m, u10, v10, sp)
are the mean of hours 00–23 UTC; tp is the sum of hours 01–24 UTC.
CONUS404 (conus404_data/)
WRF dynamical downscaling of ERA5 by NCAR, 4km resolution. Each file holds
T2, TD2, U10, V10, PSFC, PREC_ACC_NC, plus static fields (Z,
LAI, etc.) at daily resolution, one instantaneous snapshot at 00:00 UTC
per day (PREC_ACC_NC is WRF's own accumulated diagnostic, not an
instantaneous value at that timestamp).
Note: ERA5's daily value is a 24 hour mean/sum; CONUS404's is a single 00Z instantaneous snapshot. These are not the same point in time, by construction of the two source products.
Usage
# Just ERA5 (~6 GB)
hf download mudhil/era5-conus404-data --repo-type dataset --local-dir /discover/nobackup/hpmille1/data --include "era5_data/*.nc"
# Just CONUS404 (~2.4 TB, large)
hf download mudhil/era5-conus404-data --repo-type dataset --local-dir /discover/nobackup/hpmille1/data --include "conus404_data/*.nc"
# Both
hf download mudhil/era5-conus404-data --repo-type dataset --local-dir /discover/nobackup/hpmille1/data
Then regenerate the repo-local flat symlinks the training/eval code reads
(data/era5_{YYYY}.nc, data/conus404_yearly_{YYYY}.nc) with
data/link_data.sh
from the model repo, or point DATA_ROOT at wherever you downloaded this to.
See milhud/diffusion_downscaling_model
for the full pipeline code.
- Downloads last month
- 81