GOES-R ABI — download a CONUS granule (live)¶

This notebook fetches a real GOES-East ABI granule from the anonymous noaa-goes19 bucket via the earthlens.goes backend, then decodes one infrared band with pyramids and maps it.

earthlens.goes ships the raw NetCDF and does not decode it — reading and reprojecting the geostationary grid is pyramids' job downstream. This notebook shows that hand-off end to end.

Setup¶

Imports and a notebook-relative output directory. GOES CONUS imagery updates every 5 minutes with a short latency, so we pick a window a few hours in the past to be sure the granules are already published.

In [1]:

from pathlib import Path

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from earthlens import EarthLens

out_dir = Path("goes_out")
out_dir.mkdir(exist_ok=True)

from pathlib import Path import numpy as np import pandas as pd import matplotlib.pyplot as plt from earthlens import EarthLens out_dir = Path("goes_out") out_dir.mkdir(exist_ok=True)

In [2]:

now = pd.Timestamp.utcnow().tz_localize(None).floor("min")
start = now - pd.Timedelta(hours=3)
end = start + pd.Timedelta(minutes=6)          # one CONUS scan (5-min cadence)
start, end

now = pd.Timestamp.utcnow().tz_localize(None).floor("min") start = now - pd.Timedelta(hours=3) end = start + pd.Timedelta(minutes=6) # one CONUS scan (5-min cadence) start, end

C:\Users\main\AppData\Local\Temp\ipykernel_17764\2842703603.py:1: Pandas4Warning: Timestamp.utcnow is deprecated and will be removed in a future version. Use Timestamp.now('UTC') instead.
  now = pd.Timestamp.utcnow().tz_localize(None).floor("min")

Out[2]:

(Timestamp('2026-07-04 19:58:00'), Timestamp('2026-07-04 20:04:00'))

Download¶

EarthLens("goes", ...).download() lists the hour prefix, keeps the granules whose scan-start lands in the window, and downloads them whole. It returns the list[pathlib.Path] of raw NetCDF files.

In [3]:

lens = EarthLens(
    "goes",
    dataset="abi-l2-mcmip",                    # Cloud & Moisture Imagery (16 bands)
    satellite="east",                          # -> noaa-goes19
    domain="C",                                # CONUS
    start=start.strftime("%Y-%m-%d %H:%M"),
    end=end.strftime("%Y-%m-%d %H:%M"),
    fmt="%Y-%m-%d %H:%M",
    lat_lim=[20, 50], lon_lim=[-130, -60],
    path=str(out_dir),
)
paths = lens.download()
[p.name for p in paths]

lens = EarthLens( "goes", dataset="abi-l2-mcmip", # Cloud & Moisture Imagery (16 bands) satellite="east", # -> noaa-goes19 domain="C", # CONUS start=start.strftime("%Y-%m-%d %H:%M"), end=end.strftime("%Y-%m-%d %H:%M"), fmt="%Y-%m-%d %H:%M", lat_lim=[20, 50], lon_lim=[-130, -60], path=str(out_dir), ) paths = lens.download() [p.name for p in paths]

2026-07-05 00:58:51.767 | INFO     | earthlens.goes.backend:_search:484 - goes: planned 1 granule(s) for ABI-L2-MCMIPC on noaa-goes19

goes:   0%|          | 0/1 [00:00<?, ?granule/s]

goes: 100%|██████████| 1/1 [00:02<00:00,  2.55s/granule]

goes: 100%|██████████| 1/1 [00:02<00:00,  2.55s/granule]

Out[3]:

['OR_ABI-L2-MCMIPC-M6_G19_s20261852001182_e20261852003555_c20261852004251.nc']

The filename encodes everything: OR_ABI-L2-MCMIPC-M6_G19_s…_e…_c….nc — the product (MCMIPC), the scan mode (M6), the satellite (G19), and the scan start / end / created timestamps. One CONUS file (~58 MB) carries all 16 ABI bands.

In [4]:

granule = paths[0]
print("granule :", granule.name)
print("size    :", round(granule.stat().st_size / 1e6, 1), "MB")

granule = paths[0] print("granule :", granule.name) print("size :", round(granule.stat().st_size / 1e6, 1), "MB")

granule : OR_ABI-L2-MCMIPC-M6_G19_s20261852001182_e20261852003555_c20261852004251.nc
size    : 58.5 MB

Decode a band with pyramids¶

The granule is raw geostationary NetCDF. pyramids georeferences the ABI scan-angle grid from the CF goes_imager_projection grid-mapping, so we can pull the clean longwave IR window band (CMI_C13, 10.3 µm — the classic IR cloud channel) and warp it to WGS84.

In [5]:

from pyramids.netcdf import NetCDF

nc = NetCDF.read_file(str(granule))
band = nc.get_variable("CMI_C13").to_crs(4326)
image = np.asarray(band.read_array()).astype("float32")
image[image >= 65535] = np.nan                 # mask the fill value
image.shape

from pyramids.netcdf import NetCDF nc = NetCDF.read_file(str(granule)) band = nc.get_variable("CMI_C13").to_crs(4326) image = np.asarray(band.read_array()).astype("float32") image[image >= 65535] = np.nan # mask the fill value image.shape

2026-07-05 00:59:01 | INFO | pyramids.base.config | Logging is configured.

2026-07-05 00:59:03 | WARNING | pyramids.base.config.gdal | GDAL[1] Unhandled X/Y axis unit rad. SRS will ignore axis unit and be likely wrong.

2026-07-05 00:59:03 | ERROR | pyramids.base.config.gdal | GDAL[1] arrayStartIdx[0] + (count[0]-1) * arrayStep[0] >= 1500

Out[5]:

(1144, 2682)

Map it¶

Colder cloud tops are bright in the inverted IR palette; the warm surface is dark. The values are the granule's packed brightness-temperature counts (lower = warmer), which is why we invert the colormap.

In [6]:

fig, ax = plt.subplots(figsize=(9, 6))
vmin, vmax = np.nanpercentile(image, [2, 98])
im = ax.imshow(image, cmap="Greys", vmin=vmin, vmax=vmax)
ax.set_title(f"GOES-19 ABI CMI_C13 (10.3 µm) — {granule.name.split('_s')[1][:11]}")
ax.set_xlabel("pixels (west→east)")
ax.set_ylabel("pixels (north→south)")
fig.colorbar(im, ax=ax, shrink=0.7, label="packed IR counts (low = warm)")
plt.tight_layout()
plt.show()

fig, ax = plt.subplots(figsize=(9, 6)) vmin, vmax = np.nanpercentile(image, [2, 98]) im = ax.imshow(image, cmap="Greys", vmin=vmin, vmax=vmax) ax.set_title(f"GOES-19 ABI CMI_C13 (10.3 µm) — {granule.name.split('_s')[1][:11]}") ax.set_xlabel("pixels (west→east)") ax.set_ylabel("pixels (north→south)") fig.colorbar(im, ax=ax, shrink=0.7, label="packed IR counts (low = warm)") plt.tight_layout() plt.show()

Takeaway¶

• EarthLens("goes", ...).download() returns raw ABI NetCDF granule paths — no decoding, no server-side subset.
• Reading / reprojecting the geostationary grid is a downstream pyramids (or satpy) step — here NetCDF.read_file(...).get_variable("CMI_C13").to_crs(4326).
• Swap domain="F" for Full Disk, domain="M1" for a 1-minute mesoscale sector, or dataset="abi-l1b-rad", variables=["C02"] for a single-channel radiance file. See the catalog explorer.