The 2025 edition of Code for Earth is already in full swing, and we’re kicking off a series of short articles that highlight the stories, partnerships, and innovations behind this year’s projects. First up: a closer look at our collaboration with Helmholtz-Zentrum Hereon, a leading German research centre focused on environmental science and modelling.

From Regional Modelling to Global Scale Impact

This year’s challenge with Hereon builds directly on a successful 2024 project, where participants used machine learning to downscale regional air pollution data. That early effort revealed just how promising AI-driven methods can be in translating large-scale atmospheric data into meaningful insights at the urban level.

In 2025, the idea has grown in ambition:

Can we downscale global CAMS air quality fields over Europe and apply the models to other regions — like the US or China — while keeping results explainable, robust and scalable?

Mentors from CAMS (Copernicus Atmosphere Monitoring Service) and Helmholtz Zentrum Hereon jointly designed a challenge that pushes the boundaries of machine learning in environmental science. It calls for scalable, transferable models that can generate higher-resolution air quality data from global sources — which could prove quite impactful for regions where no detailed forecasting services currently exist.

Machine Learning for Cleaner Air

This challenge invited teams to submit a proposal experimenting with cutting-edge machine learning approaches, from Convolutional Neural Networks and Visual Transformers to Fourier Neural Operators. The emphasis is not just on innovation, but on scientific rigour — participants are using FAIRMODE metrics to evaluate performance, ensuring the work remains grounded in real-world standards.

The long-term ambition? To support better air quality forecasts in underserved parts of the world by leveraging global models and AI.

Why It Matters

The need for timely, high-resolution air pollution data is growing — and traditional modelling approaches (like chemical transport models) are often too computationally demanding to meet that need at scale. Machine learning offers a new path forward, and this challenge is a testbed for that future.

As Martin Ramacher from Hereon put it:

“We wanted a challenge that reflects a real-world need: scalable, explainable models that use the best available data and modern AI techniques. And we’re thrilled to see the diversity and quality of the proposals we received.”

About Hereon

Helmholtz-Zentrum Hereon is part of the Helmholtz Association and one of Germany’s leading institutes for environmental research. Their work spans coastal systems, urban environments, Earth observation, and machine learning. They are committed to open science, reproducibility, and cross-sector collaboration — making them a perfect partner for Code for Earth.

Stay tuned for more updates as teams continue to tackle this exciting challenge, and as we spotlight more stories from the Code for Earth community.

To learn more about Helmhotz-Zentrum Hereon, please read this article.