STAT41130 – Artificial Intelligence for Weather and Climate

Author: Andrew Parnell
Institution: University College Dublin
Inspired by: AI by Hand – Tom Yeh
Official GitHub: https://github.com/andrewcparnell/STAT41130

---

🧭 Overview

STAT41130: AI for Weather and Climate is a hands-on course designed to introduce students to machine learning methods for weather and climate applications.
It combines manual “AI by Hand” exercises, Python coding, and real-world datasets to build understanding from basic linear regression through to modern neural networks and the ECMWF Anemoi framework.

The course bridges meteorology and artificial intelligence — ideal for students from both backgrounds who want to explore AI methods for forecasting, modelling, and data-driven science.

---

🎯 Aims

• Understand how neural networks extend traditional linear regression.
• Learn key concepts of forward and backward propagation.
• Implement and train models using PyTorch.
• Explore deep learning architectures: CNNs, RNNs, Transformers, and GNNs.
• Apply AI models to weather and climate data using the Anemoi framework from ECMWF.

---

🧩 Course Structure

Format:

• 4 + 4 days intensive format
• Morning: ~2 hours of lectures
• Midday: guided coding session
• Afternoon: 2–3 hour group projects and presentations

Content progression:

1. Linear Regression & Neural Networks
2. Deep Learning Fundamentals
3. Convolutional Neural Networks
4. Recurrent Neural Networks
5. Transformers
6. Graph Neural Networks
7. Probabilistic Forecasting
8. Anemoi Ecosystem: Graphs, Models, and Training

---

🧱 Repository Contents

Folder	Description
/slides	Lecture slides in PowerPoint format (e.g., D1C1_LR_NNs.pptx) and 'by hand' worksheets
/code	Python scripts and Jupyter notebooks for coding labs
/setup	Installation instructions and requirements files for Linux and Windows
/data	Example datasets for exercises (ERA5, Anemoi samples, etc.)

---

🧠 Included Worksheets

The AI by Hand workbooks provide exercises to understand neural networks through manual calculation before coding

---

⚙️ Setup Instructions

Recommended Environment

• OS: Ubuntu (preferred) or Windows
• Editor: Visual Studio Code
• Python: 3.11

Installation

Clone the repository:

git clone https://github.com/andrewcparnell/STAT41130.git
cd STAT41130

Install dependencies:

# For Linux
pip install -r requirements_ECMWF.txt

# For Windows
pip install -r requirements_ECMWF_win.txt

For GPU support, ensure you have a CUDA-compatible version of PyTorch as per PyTorch installation guide.

---

🌦️ The Anemoi Framework

This course uses ECMWF’s Anemoi system — a modular ecosystem for machine learning in weather forecasting.

• anemoi-graphs – define graph structures for models
• anemoi-models – provides neural network architectures (GNNs, Transformers, etc.)
• anemoi-training – handles data loading, model training, and distributed computing

Students will explore these packages using simple examples before scaling to larger datasets.

References:

• Anemoi Graphs Documentation
• Anemoi Models Documentation
• Anemoi Training Documentation

---

💡 Learning Philosophy

This course emphasises:

• Understanding by doing – every lecture links to coding or hand exercises.
• Bridging theory and practice – from matrix operations to full neural networks.
• Interdisciplinary collaboration – between AI and meteorology students.
• Open-source tools – to encourage exploration beyond the classroom.

---

🧑‍💻 Contributing

If you find errors or have improvements:

1. Open an issue in this repository (intermediate)
2. Create a pull request with a fix (advanced and most helpful)
3. Or simply let the instructor know (basic)

---

📜 License

Course materials © 2025 Andrew Parnell, University College Dublin.
Anemoi packages © ECMWF, licensed under Apache 2.0.