Whole monthly station records, one file per station

Saved results from a run against the live service; the notebook records when it ran and the checksums of what it read. Run it yourself.

The Global Summary of the Month is NOAA NCEI's set of monthly station summaries derived from GHCN-Daily: means, extremes, totals, and day counts for tens of thousands of stations. Besides its query service, NCEI publishes one static CSV per station holding that station's whole monthly record, every element, in metric units. A file is one plain download, which makes it the fast way to get long records for many stations.

This walkthrough pulls the complete files for two snowbelt airports, Pittsburgh and Buffalo, and totals their snowfall winter by winter since 1950. It needs usdata[pandas] and matplotlib, and downloads about 1.3 MB.

from datetime import UTC, datetime
from pathlib import Path

import matplotlib.pyplot as plt
import pandas as pd

import usdata
from usdata import cite_lockfile, pull, verify

# One figure style for every usdata notebook, so previews look alike.
plt.rcParams.update(
    {
        "figure.figsize": (8, 4.5),
        "figure.dpi": 120,
        "axes.spines.top": False,
        "axes.spines.right": False,
        "axes.grid": True,
        "grid.alpha": 0.3,
        "font.size": 10,
    }
)
manifest = Path("dataset.yaml")
print("Executed (UTC):", datetime.now(UTC).isoformat(timespec="seconds"))
print(f"usdata {usdata.__version__}; pandas {pd.__version__}")

Executed (UTC): 2026-09-29T03:43:03+00:00
usdata 0.32.0; pandas 3.0.6

Select

The manifest names two GHCN station ids and nothing else. A station file is always the whole record of every element, so there is no window, box, or variable to give; months and columns are chosen after opening. USW00094823 is Pittsburgh International Airport and USW00014733 is Buffalo Niagara International Airport.

print(manifest.read_text())

name: two-snowbelt-records
sources:
  - dataset: noaa:gsom-station-files
    params: {stations: [USW00094823, USW00014733]}

What arrives

The first pull downloads one file per station, named by its id, and writes dataset.lock.json beside the manifest, pinning each file's checksum separately. NCEI rewrites a station's file as its summaries change, so a pin on one station can drift without touching the other.

result = pull(manifest)
for item in result.fetched:
    print(f"{item.asset.id}  {item.provenance.size:,} bytes  {item.provenance.source_url}")
print("Retrieved (UTC):", result.fetched[0].provenance.retrieved_at.isoformat(timespec="seconds"))

USW00014733.csv  688,743 bytes  https://www.ncei.noaa.gov/data/global-summary-of-the-month/access/USW00014733.csv
USW00094823.csv  605,617 bytes  https://www.ncei.noaa.gov/data/global-summary-of-the-month/access/USW00094823.csv
Retrieved (UTC): 2026-09-29T03:43:04+00:00

Open

open() returns a pandas DataFrame, one row per month from the station's first summarised month to its last. After the station, date, position, elevation, and name come the elements, each followed by its _ATTRIBUTES column. Which elements appear depends on the station. SNOW is the month's snowfall in millimetres, and a blank means NCEI could not compute it: more than five days missing or flagged. The attribute is DaysMissing,Measurement,Quality,Source, so its first field counts the days that were missing.

AIRPORTS = {"USW00094823": "Pittsburgh", "USW00014733": "Buffalo"}
frames = {}
for item in result.fetched:
    frame = item.open()
    station = frame["STATION"].iloc[0]
    frames[AIRPORTS[station]] = frame
    print(
        f"{station} {frame['NAME'].iloc[0]}: {len(frame)} months, "
        f"{frame['DATE'].iloc[0]} to {frame['DATE'].iloc[-1]}, {len(frame.columns)} columns"
    )
print("Properties:", result.fetched[0].asset.properties)
frames["Pittsburgh"].set_index("DATE").loc["1997-11":"1998-03", ["SNOW", "SNOW_ATTRIBUTES"]]

USW00014733 BUFFALO NIAGARA INTERNATIONAL AIRPOR, NY US: 1061 months, 1938-05 to 2026-09, 110 columns
USW00094823 PITTSBURGH INTERNATIONAL AIRPORT, PA US: 945 months, 1948-01 to 2026-09, 106 columns
Properties: {'units': 'metric'}

	SNOW	SNOW_ATTRIBUTES
DATE
1997-11	64.0	,,,Z
1997-12	320.0	,,,0
1998-01	45.0	,,,0
1998-02	64.0	,,,0
1998-03	125.0	,,,0

A winter here is November to March, named by the year it starts, and it counts only when all five months have a value, so a gap never reads as a snow drought.

def winter_totals(frame: pd.DataFrame) -> pd.Series:
    """Snowfall per November-March winter, where all five months have a value."""
    months = frame.assign(
        year=frame["DATE"].str[:4].astype(int), month=frame["DATE"].str[5:].astype(int)
    )
    months = months[months["month"].isin([11, 12, 1, 2, 3])]
    winter = months["year"] - (months["month"] <= 3)
    grouped = months.groupby(winter)["SNOW"]
    return grouped.sum()[grouped.count() == 5].loc[1950:2025]


winters = pd.DataFrame({name: winter_totals(frame) / 10 for name, frame in frames.items()})
print(winters.describe().round(0).loc[["count", "mean", "min", "max"]].to_string())

       Buffalo  Pittsburgh
count     76.0        76.0
mean     231.0       106.0
min       91.0        39.0
max      500.0       197.0

A first look

fig, ax = plt.subplots(layout="constrained")
for name, colour in [("Buffalo", "#2166ac"), ("Pittsburgh", "#b2182b")]:
    series = winters[name]
    ax.plot(
        series.index, series, marker="o", markersize=2.5, linewidth=0.9, color=colour, label=name
    )
ax.set_xlabel("Winter (November to March), named by its first year")
ax.set_ylabel("Snowfall (cm)")
ax.set_title("Winter snowfall at two snowbelt airports, 1950-51 to 2025-26")
ax.legend(frameon=False)
plt.show()

Saved plot from Whole monthly station records, one file per station

for name, series in winters.items():
    top = series.idxmax()
    print(
        f"{name}: snowiest winter {top}-{str(top + 1)[2:]}, "
        f"{series.max():.0f} cm; median {series.median():.0f} cm"
    )

Buffalo: snowiest winter 1976-77, 500 cm; median 225 cm
Pittsburgh: snowiest winter 2009-10, 197 cm; median 100 cm

Buffalo, downwind of Lake Erie, gets roughly twice Pittsburgh's snow in a typical winter, and its lake-effect seasons swing much further from one year to the next.

Pin and cite

verify checks both cached files against the lockfile's checksums. The citation below is what a methods section needs, and usdata cite dataset.yaml prints the same.

assert verify(manifest) == []
for citation in cite_lockfile(manifest):
    print(citation.as_text())

noaa:gsom-station-files
  Lawrimore, Jay H.; Ray, Ron; Applequist, Scott; Korzeniewski, Bryant; Menne, Matthew J. (2016): Global Summary of the Month (GSOM), Version 1. NOAA National Centers for Environmental Information. https://doi.org/10.7289/V5QV3JJ5
  homepage: https://www.ncei.noaa.gov/data/global-summary-of-the-month/
  license: US Government Work (public domain)
  terms: https://www.ncei.noaa.gov/metadata/geoportal/rest/metadata/item/gov.noaa.ncdc:C00946/html
  retrieved: 2026-09-29; 2 checksummed assets (1,294,360 bytes) pinned by usdata 0.32.0
  sources: 1

What was awkward

A file holds every element, so two stations' snowfall costs about 1.3 MB of mostly other columns; there is no way to ask for SNOW alone.
A station's span is only known after the download: the asset's time is the archive's bound from 1763, not the station's own first and last month.
Winters straddle calendar years, so totalling one takes a few lines of date arithmetic, and a blank month has to be treated as missing rather than as zero.
NCEI's NAME column is cut at 30 characters ("BUFFALO NIAGARA INTERNATIONAL AIRPOR"), so labels come from a lookup of station ids instead.
The attribute column packs four flags into one comma-separated string, so reading the days-missing count means splitting it yourself.

Whole monthly station records, one file per station

At a glance

Quick start

Walkthrough

Select

What arrives

Open

A first look

Pin and cite

What was awkward

Used in these studies

How have strong El Niño winters changed U.S. snowfall, and what does that suggest for 2026–27?

Reference