GeoFetch builds local Zarr stores from remote weather and geoscience datasets. You describe what you want as a small pipeline — a data source, optional selections, and a destination — and GeoFetch downloads only the pieces it needs and writes them into a single store with labelled dimensions.
Source | Select / Rechunk | Store
Building a pipeline does no I/O; execute is the one call that downloads and
writes.
using Pkg
Pkg.add(url="https://github.com/RallypointOne/GeoFetch.jl")The registered v0.1.0 predates the pipeline API described here, so install from the repository for now.
| Preset | Dataset | Access | Auth |
|---|---|---|---|
OISST |
NOAA Optimum Interpolation SST v2.1, daily, 0.25° global | NetCDF files over HTTPS (NCEI) | none |
GFS |
NOAA Global Forecast System forecasts, 0.25° / 0.5° / 1° global | GRIB2 messages by byte range (AWS) | none |
ERA5 |
ECMWF ERA5 reanalysis | not implemented yet | — |
Three days of sea surface temperature over the North Atlantic. OISST longitudes run from 0 to 360, so 280..350 is 80°W..10°W.
using GeoFetch, Dates
pipeline = OISST(start=Date(2025, 1, 1), stop=Date(2025, 1, 3)) |
Select(vars=:sst, lon=Between(280, 350), lat=Between(20, 60)) |
Rechunk(time=3) |
Store("oisst.zarr")
execute(pipeline)One forecast cycle at three lead times over Europe: 2 m temperature, accumulated precipitation, and temperature on two pressure levels. Each GFS file is about 540 MB, but only the selected GRIB messages are downloaded (under 10 MB here).
using GeoFetch, Dates
pipeline = GFS(start=Date(2026, 11, 8), stop=Date(2026, 11, 8), cycles=[0], steps=[0, 6, 12]) |
Select(vars=[:t2m, :apcp, :t], level=[850, 500], lon=Between(0, 30), lat=Between(35, 60)) |
Store("gfs.zarr")
execute(pipeline)GFS data has dimensions lon, lat, level (hPa), step (lead time in hours)
and time (forecast initialization). See the GFS docstring for the variable
list. Without a Select(vars=...) every variable on every level is fetched.
Stores are plain Zarr (v3 by default, Store(path; zarr_version=2) for v2) and
open with any Zarr reader:
using Zarr, Dates
g = zopen("gfs.zarr")
t2m = g["t2m"][:, :, :, :] # (lon, lat, step, time)
lon = g["lon"][:]
time = DateTime(1970) .+ Second.(g["time"][:])Time coordinates are stored as integers with CF units attributes
(seconds since 1970-01-01T00:00:00 for DateTime, days since 1970-01-01 for
Date).
Select takes one keyword per dimension, plus vars:
Select(vars=:sst) # one variable
Select(vars=[:t2m, :u10, :v10]) # several
Select(lat=Between(30, 60)) # closed interval
Select(level=[850, 500]) # explicit values
Select(time=Date(2025, 1, 1):Day(1):Date(2025, 1, 5)) # a rangeSelections are applied before anything is downloaded, so files, variables and levels that are not selected are never fetched.
Runnable scripts for each source are in examples/:
julia --project=. examples/oisst.jl
julia --project=. examples/gfs.jl