The zarr-eosdis-store library allows NASA EOSDIS Collections to be accessed efficiently by the Zarr Python library, provided they have a sidecar DMR++ metadata file generated.
This module requires Python 3.8 or greater:
$ python --version Python 3.8.2
Install from PyPI:
$ pip install zarr-eosdis-store
To install the latest development version:
$ pip install pip install git+https://github.com/nasa/zarr-eosdis-store.git@main#egg=zarr-eosdis-store
To access EOSDIS data, you need to sign in with a free NASA Earthdata Login account, which you can obtain at https://urs.earthdata.nasa.gov/.
Once you have an account, you will need to add your credentials to your ~/.netrc
file:
machine urs.earthdata.nasa.gov login YOUR_USERNAME password YOUR_PASSWORD
If you are accessing test data, you will need to use an account from the Earthdata Login test system at
https://uat.urs.earthdata.nasa.gov/ instead, adding a corresponding line to your ~/.netrc
file:
machine uat.urs.earthdata.nasa.gov login YOUR_USERNAME password YOUR_PASSWORD
To use the library, simply instantiate eosdis_store.EosdisStore
with the URL to the data file you would
like to access, pass it to the Zarr library as you would with any other store, and use the Zarr API as with any
other read-only Zarr file. Note: the URL to the data file will typically end with an HDF5 or NetCDF4 extension,
not .zarr.
from eosdis_store import EosdisStore
import zarr
# Assumes you have set up .netrc with your Earthdata Login information
f = zarr.open(EosdisStore('https://example.com/your/data/file.nc4'))
# Read metadata and data from f using the Zarr API
print(f['parameter_name'][0:0:0])
If the data has _FillValue (to flag nodata), scale_factor, or add_offset set (defined in metadata using CF-conventions) they can be retrieved from the parameter attributes.
import numpy as np
scale_factor = f['parameter_name].scale_factor
add_offset = f['parameter_name].add_offset
nodata = f['parameter_name]._FillValue
arr = f['parameter_name'][] * scale_factor + add_offset
nodata_locs = np.where(arr == nodata)
A better way to handle these is to use XArray. Rather than reading the data immediately when a slice is requested, XArray defers the read until the data is actually accessed. With the Zarr backend to XArray, the scale and offset can be set so that when the data is accessed it will apply those values. This is more efficient if the data is going to be used in other operations.
The scale_factor and get_offset will be used if specified in the NetCDF/HDF5 file.
import xarray
store = EosdisStore('https://example.com/your/data/file.nc4')
f = xarray.open_zarr(store)
# the data is not read yet
xa = f['parameter_name'][<slice>]
# convert to numpy array, data is read
arr = xa.values
The resulting array will have had scale and offset applied, and any element that is equal to the _FillValue attribute will be set to numpy nan. To use XArray without apply the scale and offset or setting the nodata to nan, supply the mask_and_scale keyword to xarray.open_zarr to False:
store = EosdisStore('https://example.com/your/data/file.nc4')
f = xarray.open_zarr(store, mask_and_scale=False)
We make use of a technique to read NetCDF4 and some HDF5 files that was prototyped by The HDF Group and USGS, described here.
To allow the technique to work with EOSDIS data, we have extended it and optimized access in the following key ways:
- The
EosdisStore
reads a DMR++ file generated by OPeNDAP to present its metadata and determine byte offsets to the Zarr library. By reusing these, we avoid needing to generate new metadata sidecar files to support new data. - The store uses HTTPS and authenticates with a
.netrc
entry, rather than the S3 API, making it compatible with EOSDIS access patterns and requirements - The store caches redirect URLs for a period of time set by the Cache-Control header. Doing this avoids the overhead of repeated redirects when accessing parts of files.
- The store uses a parallel API that allows it to make more efficient access optimizations:
- When the Zarr library accesses data that requires reading multiple near-sequential bytes in the file, the store combines these smaller requests into a single larger request.
- After an initial request to cache any authentication and redirect information, the store runs subsequent requests in parallel.
Clone the repository, then pip install
its dependencies:
pip install -r requirements.txt pip install -r requirements-dev.txt
To check code coverage and run tests:
coverage run -m pytest
To check coding style:
flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics
To build documentation, generated at docs/_build/html/index.html
:
cd docs && make html