Deep Learning Enhances Satellite-Based Air Quality Monitoring, Kuopio Dissertation Shows

A new doctoral dissertation from the University of Eastern Finland demonstrates how deep learning can significantly improve air quality monitoring using satellite data. Andrea Porcheddu, MSc in Technical Physics, will defend the research at the Faculty of Science, Forestry and Technology on the Kuopio campus. The work addresses a critical gap in environmental monitoring: traditional ground sensors are sparse, while satellite observations cover wide areas but require advanced interpretation to estimate ground-level pollution. Porcheddu's approach applies deep neural networks to fuse satellite measurements with meteorological and land-use data, producing more accurate and higher-resolution air quality maps. This matters because poor air quality is linked to millions of premature deaths annually, and better monitoring supports public health warnings and policy decisions. The dissertation also explores transfer learning to adapt models across regions with limited ground truth. The findings could enable near-real-time, global air quality tracking, empowering cities and researchers to respond faster to pollution events. As satellite constellations expand, such AI-driven methods will become essential for turning raw data into actionable environmental intelligence.