ERA5 reanalysis (ARCO-ERA5 and Earth Engine)

ERA5 is ECMWF’s global reanalysis: hourly fields on a 0.25° grid from 1940 to about a week ago, with the last three months served as the preliminary ERA5T. ERA5-Land re-runs the land surface at 0.1° with the same forcing. Both are the usual source of temperature and precipitation for a basin without a station.

Two routes. The default, arco-era5-gcs, is Google’s analysis-ready Zarr copy of hourly ERA5 on Cloud Storage: anonymous, no account, lazy. Everything else, ERA5-Land and the daily and monthly aggregates, lives on Earth Engine (source="gee") and needs Earth Engine credentials.

The figures put one week of 2 m temperature from both routes on one calendar axis, then map the ERA5-Land grid at its coldest hour to show what a reanalysis cell is: this 40 km box selects a single 0.25° ERA5 cell, and its temperature is that of the cell’s average elevation, not of a summit.

import geopandas as gpd
import pandas as pd
import shapely
import xarray as xr

import easysnowdata as esd

aoi = (-121.94, 46.72, -121.54, 46.99)  # Mount Rainier
week = "2023-03-01/2023-03-07"

box = esd.parse_aoi(aoi)
print(box)

AOI(bounds=(-121.9400, 46.7200, -121.5400, 46.9900), clip=True)

Where the product comes from, and what each route asks for.

for src in esd.catalog.get("era5").sources:
    print(f"{src.id:22} {src.title:45} {', '.join(src.requires) or 'no account'}")

arco-era5-gcs          ARCO-ERA5 (GCS)                               no account
gee                    Earth Engine                                  earthengine

The credential-free route: one week of hourly 2 m temperature and snow depth from ARCO-ERA5. The result is Dask-backed with time, latitude, longitude dims, and the subset holds the native cells whose centres fall in the box: at 0.25° that is a single cell here.

era5_ds = esd.climate.era5.load(aoi, week, variables=["2m_temperature", "snow_depth"])
print(era5_ds)

<xarray.Dataset> Size: 3kB
Dimensions:         (time: 168, latitude: 1, longitude: 1)
Coordinates:
  * time            (time) datetime64[ns] 1kB 2023-03-01 ... 2023-03-07T23:00:00
  * latitude        (latitude) float32 4B 46.75
  * longitude       (longitude) float32 4B -121.8
    spatial_ref     int64 8B 0
Data variables:
    2m_temperature  (time, latitude, longitude) float32 672B dask.array<chunksize=(24, 1, 1), meta=np.ndarray>
    snow_depth      (time, latitude, longitude) float32 672B dask.array<chunksize=(24, 1, 1), meta=np.ndarray>
Attributes: (12/16)
    valid_time_start:       1940-01-01
    valid_time_stop:        2026-06-30
    valid_time_stop_era5t:  2026-11-02
    last_updated:           2026-11-08 03:59:16.799835+00:00
    source:                 arco-era5-gcs
    source_id:              arco-era5-gcs
    ...                     ...
    data_citation:          Hersbach, H., et al. (2020). The ERA5 global rean...
    license:                Copernicus licence
    easysnowdata_version:   0.3.9.dev1+gd97689c2f
    doi:                    10.1002/qj.3803
    version:                ERA5
    cadence:                hourly

The same week from ERA5-Land on Earth Engine. ERA5-Land names the variable temperature_2m and resolves the box into 4 x 5 cells at 0.1°.

land_ds = esd.climate.era5.load(
    aoi, week, variables=["temperature_2m"], source="gee", version="ERA5_LAND"
)
print(land_ds)

<xarray.Dataset> Size: 15kB
Dimensions:         (time: 168, latitude: 4, longitude: 5)
Coordinates:
  * time            (time) datetime64[ms] 1kB 2023-03-01 ... 2023-03-07T23:00:00
  * latitude        (latitude) float64 32B 47.0 46.9 46.8 46.7
  * longitude       (longitude) float64 40B -121.9 -121.8 -121.7 -121.6 -121.5
    spatial_ref     int64 8B 0
Data variables:
    temperature_2m  (time, latitude, longitude) float32 13kB dask.array<chunksize=(48, 4, 5), meta=np.ndarray>
Attributes: (12/13)
    collection:            ECMWF/ERA5_LAND/HOURLY
    source:                gee
    source_id:             gee
    source_title:          Earth Engine
    source_url:            https://developers.google.com/earth-engine/dataset...
    product_id:            era5
    ...                    ...
    data_citation:         Hersbach, H., et al. (2020). The ERA5 global reana...
    license:               Copernicus licence
    easysnowdata_version:  0.3.9.dev1+gd97689c2f
    doi:                   10.1002/qj.3803
    version:               ERA5_LAND
    cadence:               hourly

Domain means in degrees Celsius, on one axis. timeseries labels the y axis from the array’s long_name and units attrs, so set them after the conversion.

def domain_mean_celsius(temperature_da):
    mean_da = (temperature_da.mean(dim=["latitude", "longitude"]) - 273.15).compute()
    mean_da.attrs = {"long_name": "2 m temperature", "units": "°C"}
    return mean_da


t2m_da = xr.concat(
    [
        domain_mean_celsius(era5_ds["2m_temperature"]),
        domain_mean_celsius(land_ds["temperature_2m"]),
    ],
    dim=pd.Index(
        ["ERA5, ARCO-ERA5 on GCS (0.25°)", "ERA5-Land, Earth Engine (0.1°)"],
        name="source",
    ),
)
ax = esd.plotting.timeseries(
    t2m_da, title="Mount Rainier AOI, domain-mean 2 m temperature"
)
ax.axhline(0, color="0.5", linestyle="--", linewidth=1)
ax.figure.tight_layout()

ERA5-Land runs a degree or two colder here because its 0.1° cells sit higher on average than the one 0.25° ERA5 cell the box selects; neither is a summit temperature. The map of the coldest hour shows the twenty ERA5-Land cells, the single ERA5 cell outlined, and the summit marked. The cells are geographic; the map is drawn on the AOI’s UTM grid at 1 km with nearest resampling, which keeps every cell a sharp block, and the outline and the summit are reprojected the same way.

coldest = pd.Timestamp(t2m_da.sel(source=t2m_da["source"][1]).idxmin("time").item())
coldest_da = (land_ds["temperature_2m"].sel(time=coldest) - 273.15).compute()
coldest_da.attrs = {"long_name": "ERA5-Land 2 m temperature", "units": "°C"}
grid = box.to_geobox(crs="utm", resolution=1000)
ax = esd.plotting.map(
    coldest_da.odc.reproject(grid, resampling="nearest"),
    cmap="coolwarm",
    title=f"ERA5-Land 2 m temperature, {coldest:%Y-%m-%d %H:%M} UTC",
)
lon, lat = float(era5_ds["longitude"].item()), float(era5_ds["latitude"].item())
cell = gpd.GeoSeries(
    [shapely.box(lon - 0.125, lat - 0.125, lon + 0.125, lat + 0.125)], crs="EPSG:4326"
).to_crs(grid.crs)
cell.boundary.plot(
    ax=ax, color="black", linewidth=2, label="the ERA5 cell the box selects (0.25°)"
)
summit = gpd.GeoSeries([shapely.Point(-121.7603, 46.8523)], crs="EPSG:4326").to_crs(
    grid.crs
)
ax.plot(
    summit.x.iloc[0],
    summit.y.iloc[0],
    marker="^",
    color="black",
    markersize=9,
    linestyle="none",
    label="summit, 4392 m",
)
ax.legend(fontsize=8, loc="upper right")
ax.figure.tight_layout()

Provenance travels with the data: which route served it, the citation and licence, and where final ERA5 ends and preliminary ERA5T begins in the ARCO store. A request that reaches past valid_time_stop gets era5t_from.

for key in (
    "source",
    "product_id",
    "license",
    "valid_time_stop",
    "valid_time_stop_era5t",
):
    print(f"{key}: {era5_ds.attrs[key]}")

source: arco-era5-gcs
product_id: era5
license: Copernicus licence
valid_time_stop: 2026-06-30
valid_time_stop_era5t: 2026-11-02