How to Use

This page describes the typical usage flow of CPFA, from preparing input data to running predictions, visualization, and evaluation.

Step 1: Prepare Input Data

CPFA expects a surface input file with the following structure:

  • File name: input_surface.npy

  • Location: download_data folder

  • Shape: (4, 721, 1440)

  • Order of variables:

    1. Mean sea level pressure (MSLP)

    2. 10m u-component of wind (U10)

    3. 10m v-component of wind (V10)

    4. 2m temperature (T2M)

  • File name: input_upper.npy

  • Location: download_data folder

  • Shape: (5, 13, 721, 1440)

  • Order of variables:

    1. z(Geopotential)

    2. q(Specific humidity)

    3. t(Temperature)

    4. u(Zonal wind)

    5. v(Meridional wind)

ERA5 Download Guide

  1. Visit the Copernicus Climate Data Store (CDS)

  2. Sign up and log in.

  3. Navigate to: input_surface.npy → ERA5 hourly data on single levels from 1940 to present input_upper.npy → ERA5 hourly data on pressure levels from 1940 to present

  4. Use the following settings:

  • input_surface.npy

    • Product type: Reanalysis

    • Variables (in this order):

      1. Mean sea level pressure

      2. 10m u-component of wind

      3. 10m v-component of wind

      4. 2m temperature

    • Year / Month / Day / Time: select the desired range

    • Geographical area: global coverage

    • Data format: NetCDF4 (experimental)

    • Download format: unarchived file

  • input_upper.npy

    • Product type: Reanalysis

    • Variables (in this order): 1. Geopotential 2. Specific humidity 3. Temperature 4. U-component of wind 5. V-component of wind

    • Pressure level: 1000hPa, 925hPa, 850hPa, 700hPa, 600hPa, 500hPa, 400hPa, 300hPa, 250hPa, 200hPa, 150hPa, 100hPa and 50hPa in the exact order

    • Year / Month / Day / Time: select the desired range

    • Geographical area: global coverage

    • Data format: NetCDF4 (experimental)

    • Download format: unarchived file

  1. Accept the terms of use, submit the form, and download the file.

  2. Place the downloaded ERA5 files in the download_data folder.

Step 2: Transformate file

Run .. code-block:: bash

python transform_nc_to_npy.py

Check whether the newly generated data has been successfully created in the input_data folder.

Step 3: Run the Prediction

With the virtual environment activated and input data prepared, move to the project root in Anaconda Prompt and run:

python prediction.py

This script performs the following tasks:

  • Loads input_surface.npy from input_data

  • Loads the appropriate Pangu-Weather ONNX model (e.g. 1, 3, 6, or 24-hour)

  • Executes the forward prediction

  • Stores prediction results in the output_data folder

Step 4: Visualization

After predictions are generated, you can visualize the results using:

python visualization.py

This script typically reads data from output_data and produces figures such as:

  • Global maps of temperature or pressure

  • Spatial plots of wind components

  • Time series at specific grid points (depending on implementation)

The plots are usually saved as image files (e.g. PNG)

Step 5: Evaluation

To evaluate model performance against ERA5 data, run:

python evaluation.py

This script:

  • Loads CPFA prediction outputs

  • Loads reference ERA5 data

  • Computes basic verification metrics (for example, mean error or other statistics depending on the implementation)

  • Optionally saves summary tables or plots