google / weather-tools

Tools to make weather data accessible and useful.

weather-tools

Apache Beam pipelines to make weather data accessible and useful.

Introduction

This project contributes a series of command-line tools to make common data engineering tasks easier for researchers in climate and weather. These solutions were born out of the need to improve repeated work performed by research teams across Alphabet.

The first tool created was the weather downloader (weather-dl). This makes it easier to ingest data from the European Center for Medium Range Forecasts (ECMWF). weather-dl enables users to describe very specifically what data they'd like to ingest from ECMWF's catalogs. It also offers them control over how to parallelize requests, empowering users to retrieve data efficiently. Downloads are driven from a configuration file, which can be reviewed (and version-controlled) independently of pipeline or analysis code.

We also provide two additional tools to aid climate and weather researchers: the weather mover (weather-mv) and the weather splitter (weather-sp). These CLIs are still in their alpha stages of development. Yet, they have been used for production workflows for several partner teams.

We created the weather mover (weather-mv) to load geospatial data from cloud buckets into Google BigQuery. This enables rapid exploratory analysis and visualization of weather data: From BigQuery, scientists can load arbitrary climate data fields into a Pandas or XArray dataframe via a simple SQL query.

The weather splitter (weather-sp) helps normalize how archival weather data is stored in cloud buckets: Whether you're trying to merge two datasets with overlapping variables — or, you simply need to open Grib data from XArray, it's really useful to split datasets into their component variables.

Installing

It is currently recommended that you create a local python environment (with Anaconda) and install the sources as follows:

conda env create --name weather-tools --file=environment.yml
conda activate weather-tools

Note: Due to its use of 3rd-party binary dependencies such as GDAL and MetView, weather-tools is transitioning from PyPi to Conda for its main release channel. The instructions above are a temporary workaround before our Conda-forge release.

From here, you can use the weather-* tools from your python environment. Currently, the following tools are available:

• ⛈ weather-dl (beta) – Download weather data (namely, from ECMWF's API).
• ⛅️ weather-mv (alpha) – Load weather data into analytics engines, like BigQuery.
• 🌪 weather-sp (alpha) – Split weather data by arbitrary dimensions.

Quickstart

In this tutorial, we will download the Era 5 pressure level dataset and ingest it into Google BigQuery using weather-dl and weather-mv, respectively.

Prerequisites

1. Register for a license from ECMWF's Copernicus (CDS) API.
2. Install your license by copying your API url & key from this page to a new file $HOME/.cdsapirc.¹ The file should look like this:

   url: https://cds.climate.copernicus.eu/api/v2
   key: <YOUR_USER_ID>:<YOUR_API_KEY>
   

3. If you do not already have a Google Cloud project, create one by following these steps. If you are working on an existing project, make sure your user has the BigQuery Admin role. To learn more about granting IAM roles to users in Google Cloud, visit the official docs.
4. Create an empty BigQuery Dataset. This can be done using the Google Cloud Console or via the bq CLI tool. For example:

   bq mk --project_id=$PROJECT_ID $DATASET_ID

5. Follow these steps to create a bucket for staging temporary files in Google Cloud Storage.

Steps

For the purpose of this tutorial, we will use your local machine to run the data pipelines. Note that all weather-tools can also be run in Cloud Dataflow which is easier to scale and fully managed.

1. Use weather-dl to download the Era 5 pressure level dataset.

   weather-dl configs/era5_example_config_local_run.cfg \
      --local-run # Use the local machine

   Recommendation: Pass the -d, --dry-run flag to any of these commands to preview the effects.

   NOTE: By default, local downloads are saved to the ./local_run directory unless another file system is specified. The recommended output location for weather-dl is Cloud Storage. The source and destination of the download are configured using the .cfg configuration file which is passed to the command. To learn more about this configuration file's format and features, see this reference. To learn more about the weather-dl command, visit here.

2. (optional) Split your downloaded dataset up with weather-sp:

    weather-sp --input-pattern "./local_run/era5-*.nc" \
       --output-dir "split_data" 

   Visit the weather-sp docs for more information.

3. Use weather-mv to ingest the downloaded data into BigQuery, in a structured format.

   weather-mv bigquery --uris "./local_run/**.nc" \ # or "./split_data/**.nc" if weather-sp is used
      --output_table "$PROJECT.$DATASET_ID.$TABLE_ID" \ # The path to the destination BigQuery table
      --temp_location "gs://$BUCKET/tmp" \  # Needed for stage temporary files before writing to BigQuery
      --direct_num_workers 2

   See these docs for more about the weather-mv command.

That's it! After the pipeline is completed, you should be able to query the ingested dataset in BigQuery SQL workspace and analyze it using BigQuery ML.

Contributing

The weather tools are under active development, and contributions are welcome! Please check out our guide to get started.

License

This is not an official Google product.

Copyright 2021 Google LLC

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Footnotes

1. Note that you need to be logged in for the CDS API page to actually show your user ID and API key. Otherwise, it will display a placeholder, which is confusing to some users. ↩