stactools-noaa-hrrr

• Name: noaa-hrrr
• Package: stactools.noaa_hrrr
• stactools-noaa-hrrr on PyPI
• Owner: @hrodmn
• Dataset homepage
• STAC extensions used:
  • forecast
  • item-assets
  • datacube
• Extra fields:
  • noaa-hrrr:forecast_cycle_type: either standard (18-hour) or extended (48-hour)
  • noaa-hrrr:region: either conus or alaska
• Browse the example in human-readable form
• Browse a notebook demonstrating the example item and collection

This package can be used to generate STAC metadata for the NOAA High Resolution Rapid Refresh (HRRR) atmospheric forecast dataset.

The data are uploaded to cloud storage in AWS, Azure, and Google so you can pick which cloud provider you want to use for the grib and index hrefs using the cloud_provider argument to the functions in stactools.noaa_hrrr.stac.

Background

The NOAA HRRR dataset is a continuously updated atmospheric forecast data product.

Data structure

• There are two regions: CONUS and Alaska
• Every hour, new hourly forecasts are generated for many atmospheric attributes for each region
  • All hours (00-23) get an 18-hour forecast in the conus region
  • Forecasts are generated every three hours (00, 03, 06, etc) in the alaska region
  • On hours 00, 06, 12, 18 a 48-hour forecast is generated
  • One of the products (subh) gets 15 minute forecasts (four per hour per attribute), but the sub-hourly forecasts are stored as layers within a single GRIB2 file for the forecast hour rather than in separate files.
• The forecasts are broken up into 4 products (sfc, prs, nat, subh),
• Each GRIB2 file has hundreds to thousands of variables
• Each .grib2 file is accompanied by a .grib2.idx which has variable-level metadata including the starting byte for the data in that variable (useful for making range requests instead of reading the entire file) and some other descriptive metadata

Summary of Considerations for Organizing STAC Metadata

After extensive discussions, we decided to organize the STAC metadata with the following structure:

1. Collections: Separate collections for each region-product combination

   • regions: conus and alaska
   • products: sfc, prs, nat, and subh

2. Items: Each GRIB file in the archive is represented as an item with two assets:

   • "grib": Contains the actual data.
   • "index": The .grib2.idx sidecar file.

   Each GRIB file contains the forecasts for all of a product's variables for a particular forecast hour from a reference time, so you need to combine data from multiple items to construct a time series for a forecast.

3. grib:messages: Within each "grib" asset, a grib:messages property describes each layer's GRIB message index within the GRIB2 file. This enables applications to access specific parts of the GRIB2 files without downloading the entire file.

4. grib:layer_definitions: This field is meant to be populated at in the item-assets definitions at the collection level. It describes the variable name, units, and other relevant properties for the each layer within the GRIB file.

• We intend to propose a GRIB STAC extension with the grib:messages and grib:layer_definitions properties for describing the grib message indexes of the variables within the GRIB files.
• The layer-level metadata is worth storing in STAC because you can construct URIs for specific layers that GDAL can read using the VRT URI syntaxvrt://:
  • vrt:///vsicurl/{grib_href}?bands={grib_message}, where grib_message is the index of the layer within the GRIB2 file.
    • under the hood, GDAL's vrt driver is reading the sidecar .grib2.idx file and translating it into a /vsisubfile URI.

Advantages

• Applications can use grib:layer_definitions and grib:messages to create layer-specific data sets, facilitating efficient data handling.
• Splitting by region and product allows defining coherent collection-level datacube metadata, enhancing accessibility.

Disadvantages

- Storing layer-level metadata like byte ranges in the STAC metadata bloats the STAC items because there are hundreds to thousands of layers in each GRIB2 file.

For more details, please refer to the related issue discussion and pull requests #3 and #6.

STAC examples

• Collection
• Item

Python usage example

• Check out the example notebook for examples of how to create STAC metadata and how to use STAC items with grib:layers metadata to load the data into xarray.

Installation

Install stactools-noaa-hrrr with pip:

pip install stactools-noaa-hrrr

Command-line usage

To create a collection object:

stac noaahrrr create-collection {region} {product} {cloud_provider} {destination_file}

e.g.

stac noaahrrr create-collection conus sfc azure example-collection.json

To create an item:

stac noaahrrr create-item \
  {region} \
  {product} \
  {cloud_provider} \
  {reference_datetime} \
  {forecast_hour} \
  {destination_file}

e.g.

stac noaahrrr create-item conus sfc azure 2024-05-01T12 10 example-item.json

To create all items for a date range:

stac noaahrrr create-item-collection \
  {region} \
  {product} \
  {cloud_provider} \
  {start_date} \
  {end_date} \
  {destination_folder}

e.g.

stac noaahrrr create-item-collection conus sfc azure 2024-05-01 2024-05-31 /tmp/items

Docker

You can launch a jupyterhub server in a docker container with all of the dependencies installed using these commands:

docker/build
docker/jupyter

Use stac noaahrrr --help to see all subcommands and options.

Contributing

We use pre-commit to check any changes. To set up your development environment, install uv:

uv sync
uv run pre-commit install

To check all files:

uv run pre-commit run --all-files

To run the tests:

uv run pytest -vv

If you've updated the STAC metadata output, update the examples:

uv run scripts/update-examples