The ever-growing number of satellites provides an unprecedented upportunity to address grand environmental challenges, such as global warming. Effectively leveraging such data requires end-to-end modeling of satellite imagery, heavily relying on machine learning techniques. However, configuring advanced machine learning pipelines is a complex and inaccessible task for domain experts. Automated Machine Learning (AutoML) is a growing research area focused on automatic configuration and design of machine learning systems.

This project aims at advancing AutoML systems targeting machine learning tasks based on Earth Observation (EO) Datasets. We address a range of different data-driven modelling tasks through effectively fusing multi-modal spatio-temporal data sources. In this project, our primary focus is on developing neural architecture search systems—a class of AutoML frameworks that autonomously design high-performance neural networks, specifically tailored for EO data analysis.

Our focus is specifically on tasks based on satellite data that observe atmospheric parameters, such as detection of methane plumes. Methane is a strong greenhouse gas and identifying its sources is an important step to reduce methane emissions and combat climate change. This work is being conducted in close collaboration with the SRON Methane Group and the European Space Agency.

• atmosphere
• earth observation

Share on Facebook Share by Bluesky Share on LinkedIn Share by WhatsApp Share by Mastodon