Open NWP forecasts — catalog explorer (no network)¶

Explore the bundled NWP model catalog offline: the models, their providers / fetch backends, and one model's forecast axis + bands. No bucket or Herbie access — runs deterministically at docs-build time.

Setup¶

Consolidate the imports up front. earthlens.nwp.Catalog loads the bundled NWP model catalog (no network); Counter tallies the models by category and matplotlib draws the summary chart at the end.

In [ ]:

from collections import Counter

import matplotlib.pyplot as plt

from earthlens.nwp import Catalog

from collections import Counter import matplotlib.pyplot as plt from earthlens.nwp import Catalog

Load the catalog¶

Instantiate the catalog and count how many NWP models it ships with.

In [ ]:

cat = Catalog()
print(f'NWP models: {len(cat.datasets)}')

cat = Catalog() print(f'NWP models: {len(cat.datasets)}')

Models by fetch backend¶

Each model is fetched through one of a handful of backends (herbie / ecmwf-opendata / direct-https / meteofrance-api). Group the models by backend to see how the catalog splits across them.

In [ ]:

# Models grouped by fetch backend (herbie / ecmwf-opendata / direct-https / meteofrance-api)
dict(sorted(Counter(m.backend for m in cat.datasets.values()).items()))

# Models grouped by fetch backend (herbie / ecmwf-opendata / direct-https / meteofrance-api) dict(sorted(Counter(m.backend for m in cat.datasets.values()).items()))

Models by provider¶

The same models, grouped instead by provider — the upstream organisation (NOAA NODD, DWD, ECMWF, …) that publishes each forecast.

In [ ]:

# Models grouped by provider
dict(sorted(Counter(m.provider for m in cat.datasets.values()).items()))

# Models grouped by provider dict(sorted(Counter(m.provider for m in cat.datasets.values()).items()))

One model's forecast axis and bands¶

Pull a single model (gfs) and inspect its forecast geometry — the run cycles, the forecast horizon, the step cadence, and a sample of its bands.

In [ ]:

# One model's forecast axis and a few bands
gfs = cat.get_dataset('gfs')
print('provider :', gfs.provider, '| backend:', gfs.backend)
print('cycles   :', gfs.cycles_utc)
print('horizon  :', gfs.horizon_h, 'h | step cadence:', gfs.step_cadence_h, 'h')
print('bands    :', list(gfs.bands)[:6], f'... ({len(gfs.bands)} total)')

# One model's forecast axis and a few bands gfs = cat.get_dataset('gfs') print('provider :', gfs.provider, '| backend:', gfs.backend) print('cycles :', gfs.cycles_utc) print('horizon :', gfs.horizon_h, 'h | step cadence:', gfs.step_cadence_h, 'h') print('bands :', list(gfs.bands)[:6], f'... ({len(gfs.bands)} total)')

Chart: models by fetch backend¶

Finally, draw the backend tally as a bar chart for a quick visual of how the catalog is distributed across fetch backends.

In [ ]:

by_backend = Counter(m.backend for m in cat.datasets.values())

fig, ax = plt.subplots(figsize=(8, 4))
ax.bar(list(by_backend), list(by_backend.values()), color='tab:green')
ax.set(ylabel='models', title=f'{len(cat.datasets)} NWP models by fetch backend')
plt.xticks(rotation=20, ha='right')
plt.tight_layout()
plt.show()

by_backend = Counter(m.backend for m in cat.datasets.values()) fig, ax = plt.subplots(figsize=(8, 4)) ax.bar(list(by_backend), list(by_backend.values()), color='tab:green') ax.set(ylabel='models', title=f'{len(cat.datasets)} NWP models by fetch backend') plt.xticks(rotation=20, ha='right') plt.tight_layout() plt.show()