Note
Go to the end to download the full example code.
Daily snow-station archive (global_snow_networks)#
global_snow_networks pre-downloads daily SWE and snow depth for every
station across its networks whose daily record its probe has verified,
and publishes three things: a normalized inventory of every station, periodic
snow courses included; the observations as a chunked Zarr store on its GitHub
Pages site; and one CSV per daily station bundled into a single archive of
about 28 MB. All are rebuilt daily.
Three routes: source="github-pages-zarr" (default) reads the store and
fetches only the chunks a request touches — under a megabyte for one water
year of every station, or for one station’s whole record — and falls back to
the snapshot bundle if Pages cannot be read; source="github-tarball"
downloads the bundle attached to the latest snapshot release (each has a
Zenodo DOI) once into the package cache and reads every wanted CSV out of it,
which is the route for a result that should cite a fixed snapshot;
source="github-csv" fetches one CSV per station. None needs an account,
not even for the Norwegian stations. The archive holds SWE and snow depth
only.
The figures show every daily station coloured by the length of its record, one water year of peak SWE against elevation for all of them, and one station’s complete daily record.
import matplotlib.pyplot as plt
import pandas as pd
import easysnowdata as esd
print(f"easysnowdata {esd.__version__}")
easysnowdata 0.3.9.dev1+gd97689c2f
Where the product comes from.
for src in esd.catalog.get("snow-station-archive").sources:
print(f"{src.id:20} {src.title:45} {', '.join(src.requires) or 'no account'}")
github-pages-zarr global_snow_networks chunked archive (Pages) no account
github-tarball global_snow_networks snapshot bundle (latest release) no account
github-csv global_snow_networks per-station CSVs no account
The inventory is one HTTP request, and daily_or_better is the probe’s
verdict: a station counts as daily only once daily values have actually been
retrieved from it. Most records start around 1980; the oldest are the NWS
Cooperative Observer sites AWDB mirrors, with snow depth from the 1890s.
inv_gdf = esd.stations.archive.inventory(daily_only=True)
print(inv_gdf["network"].value_counts().to_string())
span = inv_gdf["latest_record_date"] - inv_gdf["earliest_record_date"]
inv_gdf["record_length"] = pd.cut(
span.dt.days / 365.25,
[0, 15, 30, 45, 200],
labels=["under 15 years", "15 to 30 years", "30 to 45 years", "over 45 years"],
)
inv_gdf["record_length"] = (
inv_gdf["record_length"].cat.add_categories("no dates").fillna("no dates")
)
oldest_gdf = inv_gdf.nsmallest(3, "earliest_record_date")
print(oldest_gdf[["name", "network", "earliest_record_date"]].to_string())
network
awdb 1219
fmi 902
slf 303
cdec 189
databc 148
alberta 84
snih 35
nve 32
yukon 17
ceaza 17
name network earliest_record_date
code
FMI-100971 Helsinki Kaisaniemi fmi 1892-05-01
354003_OR_COOP Hood River Exp Stn awdb 1896-12-01
0546_AK_COOP Barrow WSO Airport awdb 1901-09-03
Every probe-verified daily station, by length of record. The largest class is over 45 years: SNOTEL and the CCSS pillows were built out around 1980 and their daily archives begin there. An ordered categorical keeps the legend in that order. Two hundred stations, nearly all CDEC, carry the daily flag but no record dates yet.
print(inv_gdf["record_length"].value_counts().to_string())
ax = esd.plotting.points(
inv_gdf,
column="record_length",
legend_label="record length",
markersize=6,
figsize=(12, 5),
title=f"{len(inv_gdf)} daily snow stations",
)
record_length
over 45 years 896
under 15 years 818
15 to 30 years 630
30 to 45 years 602
no dates 0
/home/runner/work/easysnowdata/easysnowdata/docs/gallery/stations/plot_station_archive.py:66: GeographicAxesWarning: The data are in a geographic CRS (EPSG:4326), so one axis unit is a degree, not a metre. The axes use a latitude-corrected aspect (1/cos(13.3°) = 1.03) so shapes are not stretched; for a true equal-area grid load the product with crs='utm'.
ax = esd.plotting.points(
One water year everywhere. Each station’s series is cut to the window before the dense grid is built, so this is much cheaper than the whole archive.
obs_ds = esd.stations.archive.load(time="2023-10/2024-09")
print(obs_ds)
<xarray.Dataset> Size: 43MB
Dimensions: (station: 3074, time: 366)
Coordinates: (12/31)
* station (station) <U16 197kB '0280_AK_COOP' ... 'WWC_CA_MSNT'
network (station) <U11 135kB 'awdb' 'awdb' ... 'cdec' 'awdb'
name (station) <U79 971kB 'Anchorage Wscmo AP' ... 'West...
client (station) <U11 135kB 'awdb' 'awdb' ... 'cdec' 'awdb'
network_code (station) <U8 98kB 'COOP' 'COOP' ... 'CCSS' 'MSNT'
network_name (station) <U25 307kB '' '' '' '' '' ... '' '' '' '' ''
... ...
metadata_fetched_at (station) <U19 234kB '2026-11-07 00:00:00' ... '202...
station_id (station) <U16 197kB '0280:AK:COOP' ... 'WWC:CA:MSNT'
hemisphere (station) <U8 98kB 'northern' ... 'northern'
* time (time) datetime64[ns] 3kB 2023-10-01 ... 2024-09-30
water_year (station, time) int64 9MB 2024 2024 2024 ... 2024 2024
dowy (station, time) int64 9MB 1 2 3 4 ... 363 364 365 366
Data variables:
swe (station, time) float64 9MB nan nan nan ... 0.0 0.0
snwd (station, time) float64 9MB 0.0 0.0 0.0 ... nan nan
Attributes:
source: github-pages-zarr
source_id: github-pages-zarr
source_title: global_snow_networks chunked archive (Pages)
source_url: https://egagli.github.io/global_snow_networks/arch...
product_id: snow-station-archive
title: Daily snow-station archive (global snow networks)
data_citation: Gagliano, E. Global snow networks: a documented in...
license: Per contributing network; see each network's produ...
easysnowdata_version: 0.3.9.dev1+gd97689c2f
interval: daily
The archive passes the networks’ values through unfiltered. A few stations report a peak of ten metres or more of water in a single day, which is a sensor fault rather than snow; find them before they set the axis.
peak_da = obs_ds["swe"].max(dim="time")
suspect_da = peak_da.where(peak_da > 300, drop=True)
print(suspect_da.to_series().round(0).to_string())
station
75.93.0 1435.0
DPO_CA_MSNT 995.0
MB3_CA_MSNT 2073.0
Peak SWE against elevation, one point per station, coloured by network, with the faulty values clipped off the top. The same peak sits a kilometre lower in the maritime Cascades and coast ranges than in the Sierra Nevada or the continental interior.
fig, ax = plt.subplots(figsize=(8, 5))
for network in sorted(set(obs_ds["network"].values.tolist())):
rows_da = obs_ds["network"] == network
ax.scatter(
obs_ds["elevation_m"][rows_da],
peak_da[rows_da],
s=10,
alpha=0.6,
label=str(network),
)
ax.set_xlabel(esd.plotting.label(obs_ds["elevation_m"], name="elevation", units="m"))
ax.set_ylabel(esd.plotting.label(obs_ds["swe"], name="peak SWE, water year 2024"))
ax.set_title("Peak snow water equivalent against elevation")
ax.set_ylim(0, 300)
ax.grid(True, color="0.85", linewidth=0.6)
ax.set_axisbelow(True)
ax.legend(title="network", fontsize=8, frameon=False, loc="upper left")
fig.tight_layout()
One station’s whole record, from the per-station CSV route: a single request and no bundle. Paradise on Mount Rainier begins in October 1980, which is where most of the archive begins.
paradise_ds = esd.stations.archive.load("679_WA_SNTL", source="github-csv")
ax = esd.plotting.timeseries(
paradise_ds["swe"].squeeze("station"),
title="Paradise (1570 m): the whole daily record",
)
For anything the archive does not hold (precipitation, temperature, wind,
quality flags, sub-daily data, or today’s observation) go to the networks
themselves with easysnowdata.stations.load().