Access ERA5 data from NCAR’s Geoscience Data Exchange (GDEX) and compute anomaly - ARCO ERA-5 Interactive Visualization Cookbook

Overview¶

ERA5 data is available from NCAR’s GDEX in Analysis Ready Cloud Optimized (ARCO) zarr or virtual zarr formats (kerchunk). To learn how to create the annual zarr store used in this notebook, please see this notebook.

In this notebook,

We will read data zarr stores from NCAR’s GDEX endpoint
Compute temperature anomaly for the years 1940-2023

Prerequisites¶

Concepts	Importance	Notes
Intro to Xarray	Necessary
Intro to Intake	Necessary
Understanding of Zarr	Helpful

Time to learn: 30 minutes

Imports¶

import matplotlib.pyplot as plt
import numpy as np
import xarray as xr
import intake_esm
import intake
import pandas as pd
import cartopy.crs as ccrs  # Correct import for coordinate reference systems
import cartopy.feature as cfeature
from holoviews import opts
import geoviews as gv
import holoviews as hv
import aiohttp

Specify global variables¶

baseline_year_start = 1940
baseline_year_end   = 1949
current_year  = 2023

gdex_scratch   = "/glade/derecho/scratch/harshah"
annual_means_https= 'https://data.gdex.ucar.edu/special_projects/pythia_2024/annual_means/'

Create a Dask cluster¶

Dask Introduction¶

Dask is a solution that enables the scaling of Python libraries. It mimics popular scientific libraries such as numpy, pandas, and xarray that enables an easier path to parallel processing without having to refactor code.

There are 3 components to parallel processing with Dask: the client, the scheduler, and the workers.

The Client is best envisioned as the application that sends information to the Dask cluster. In Python applications this is handled when the client is defined with client = Client(CLUSTER_TYPE). A Dask cluster comprises of a single scheduler that manages the execution of tasks on workers. The CLUSTER_TYPE can be defined in a number of different ways.

There is LocalCluster, a cluster running on the same hardware as the application and sharing the available resources, directly in Python with dask.distributed.
In certain JupyterHubs Dask Gateway may be available and a dedicated dask cluster with its own resources can be created dynamically with dask.gateway.
On HPC systems dask_jobqueue is used to connect to the HPC Slurm and PBS job schedulers to provision resources.

The dask.distributed client python module can also be used to connect to existing clusters. A Dask Scheduler and Workers can be deployed in containers, or on Kubernetes, without using a Python function to create a dask cluster. The dask.distributed Client is configured to connect to the scheduler either by container name, or by the Kubernetes service name.

Select the Dask cluster type¶

The default will be LocalCluster as that can run on any system.

If running on a HPC computer with a PBS Scheduler, set to True. Otherwise, set to False.

USE_PBS_SCHEDULER = False

If running on Jupyter server with Dask Gateway configured, set to True. Otherwise, set to False.

USE_DASK_GATEWAY = False

Python function for a PBS Cluster

# Create a PBS cluster object
def get_pbs_cluster():
    """ Create cluster through dask_jobqueue.   
    """
    from dask_jobqueue import PBSCluster
    cluster = PBSCluster(
        job_name = 'dask-pythia-24',
        cores = 1,
        memory = '4GiB',
        processes = 1,
        local_directory = gdex_scratch + '/dask/spill',
        resource_spec = 'select=1:ncpus=1:mem=4GB',
        queue = 'casper',
        walltime = '1:00:00',
        #interface = 'ib0'
        interface = 'ext'
    )
    return cluster

Python function for a Gateway Cluster

def get_gateway_cluster():
    """ Create cluster through dask_gateway
    """
    from dask_gateway import Gateway

    gateway = Gateway()
    cluster = gateway.new_cluster()
    cluster.adapt(minimum=2, maximum=4)
    return cluster

Python function for a Local Cluster

def get_local_cluster():
    """ Create cluster using the Jupyter server's resources
    """
    from distributed import LocalCluster, performance_report
    cluster = LocalCluster()    

    cluster.scale(4)
    return cluster

Python logic to select the Dask Cluster type

This uses True/False boolean logic based on the variables set in the previous cells

# Obtain dask cluster in one of three ways

if USE_PBS_SCHEDULER:
    cluster = get_pbs_cluster()
elif USE_DASK_GATEWAY:
    cluster = get_gateway_cluster()
else:
    cluster = get_local_cluster()

# Connect to cluster
from distributed import Client
client = Client(cluster)

# Display cluster dashboard URL
cluster

/home/runner/micromamba/envs/ERA5_interactive/lib/python3.14/site-packages/distributed/node.py:195: UserWarning: Port 8787 is already in use.
Perhaps you already have a cluster running?
Hosting the HTTP server on port 35273 instead
  warnings.warn(

Open ERA5 annual means file¶

temp_2m_annual = xr.open_zarr(annual_means_https + 'temp_2m_annual_1940_2023.zarr').VAR_2T
temp_2m_annual

Fetching long content....

baseline_temp = temp_2m_annual.sel(time= \
                                   temp_2m_annual.time.dt.year.isin(range(baseline_year_start, baseline_year_end+1))).mean('time')
baseline_temp

temp_anomaly = temp_2m_annual - baseline_temp
temp_anomaly

Save anomaly¶

# %%time
# temp_anomaly.to_dataset().to_zarr(annual_means + 'temp_anomaly_wrt_1940_1950.zarr')

# temp_anomaly = xr.open_zarr(annual_means_https + 'temp_anomaly_wrt_1940_1950.zarr').VAR_2T
temp_anomaly

Plot the temperature anomaly

temp_anomaly.isel(time=83).plot()

Close the Dask Cluster¶

It’s best practice to close the Dask cluster when it’s no longer needed to free up the compute resources used.

cluster.close()