FAQ
This page provides answers to frequently asked questions about CPFA.
General
Q: What is CPFA?
A: CPFA (Climate Prediction For All) is an open-source educational tool for running, visualizing, and evaluating climate predictions using Pangu-Weather models.
Q: What is the main purpose of CPFA?
A: The main purpose is to provide a simple and transparent workflow for climate prediction experiments, particularly in educational and self-study settings.
Installation
Q: Which operating system is CPFA documented for?
A: CPFA is primarily documented for Windows with Anaconda and Python 3.9.2.
Q: Which Python version should I use?
A: Python 3.9.2 is recommended and used in the installation guide.
Q: Which libraries are required?
A: At minimum, NumPy, pandas, matplotlib, xarray, cartopy, ONNX (1.13.1), and ONNX Runtime (1.14.0).
Usage
Q: What input format does CPFA require?
A: The main input is input_surface.npy with shape
(4, 721, 1440) and variable order
[MSLP, U10, V10, T2M], stored in the input_data folder.
Q: Where do prediction outputs go?
A: Prediction outputs and derived products are stored in the
output_data folder.
Q: How do I generate visualizations?
A: After running the prediction script, run visualization.py to
generate plots.
Evaluation
Q: How do I compare predictions with ERA5 data?
A: Use the evaluation.py script. It reads prediction outputs and
ERA5 reference data, then computes basic performance metrics.
Q: What metrics are supported?
A: The exact metrics depend on the implementation of evaluation.py.
Typical metrics include differences or simple error statistics.
Contributing
Q: How can I report a bug?
A: Open an issue on the CPFA GitHub repository with detailed information about the problem and your environment.
Q: Is there a way to propose new features?
A: Yes. Create a feature request issue on GitHub and describe the motivation, expected behavior, and possible design.
Support
Q: How can I contact the maintainers?
A: You can use GitHub Issues for public questions and bug reports. For urgent or private matters, use the contact information listed on the project communication channels or README.