Polar weather prediction with Graph neural networks

📝 Description

This project is a part of a larger research project. The master project or summer job will be adapted and focused on a particular (smaller) aspect. The main goal is to find new, efficient, Graph neural networks for weather prediction. For the student project, it consists in testing Graph NN models (and creating new ones) on a small dataset and on wind prediction. The summer job aim at analysing weather data from an array of sensors over Norway (see references) using graph theory, network science and machine learning. Potential impact: better understanding of the polar climate, better prediction of polar storms and polar lows. could be highly valuable to find spots for offshore windmills, for an optimal production of energy.

Data: produced by weather models (we have some experts working on these models) / network of sensors (temperature, wind… + satellite images). Data on an irregular grid (Satellite data, sensors on buoys or inland).

📚 References:

• Weather prediction with deep learning: https://sites.google.com/view/meshgraphnets and Graphcast, Microsoft model
• Some open data for Norway: https://seklima.met.no/observations/

📨 Contact:

Benjamin Ricaud benjamin.ricaud@uit.no