The doctoral dissertation in the field of Technical Physics will be examined at the Faculty of Science, Forestry and Technology, Kuopio campus.
What is the topic of your doctoral research? Why is it important to study the topic?
The topic of my doctoral research regards air quality and its estimation using a fusion of data sources, in particular satellite data. Poor air quality leads every year to numerous deaths around the world: it is therefore important to understand which areas are the most affected so that policymakers can promptly act to reduce air pollution and save lives. Ground monitoring stations to estimate air pollution exist but are sparsely located and mainly present in developed countries. Satellite data can enable global estimates, complementing ground stations and helping to democratize air quality.
What are the key findings or observations of your doctoral research?
We developed a novel framework to estimate air quality using global model data products and satellite data. The model data is coarse in spatial resolution, but available globally and frequently in time. Satellite data has generally finer spatial resolutions and different temporal resolutions (in our case we worked with polar orbiting satellites with temporal resolution of 1 day and geostationary satellites with temporal resolution of 10 minutes). Combining them with other data sources (such as ground stations and geographical indicators) we aimed to transform coarse resolution model data into fine grained hourly maps that can help policymakers better understand air pollution in urban environments.
What are the key research methods and materials used in your doctoral research?
To achieve our results, we utilized a modern data driven approach. In particular, we considered different Neural Network architectures to combine multiple data sources (data fusion). As a first step, we developed a methodology to estimate ground level PM2.5 at polar orbiting satellite overpass using an ensemble of Multi-Layer Perceptrons. We then utilized a 3D U-Net to obtain hourly fine grained maps from these polar orbiting satellite maps, model data and ground stations data. These first two steps constitute a framework to estimate hourly PM2.5 maps on a 100 m resolution grid. Finally, as a first step to extend this framework to geostationary satellites, we combined GOES-East and GOES-West satellite data over the CONtiguous United States (CONUS) region using a Transformer architecture.
The doctoral dissertation of Andrea Porcheddu, MSc, entitled From space to street level: Deep learning-enhanced air quality monitoring using satellite data will be examined at the Faculty of Science, Forestry and Technology, Kuopio campus. The opponent will be Associate Professor Pauli Paasonen, Institute for Atmospheric and Earth System Research (INAR), and the custos will be Professor Ville Kolehmainen, University of Eastern Finland. Language of the public defence is English.
