AI and physical models for air quality monitoring at urban scale with PRISMA hyperspectral data

The challenge of air pollution and its impact on human health is a significant concern in contemporary society. The PRIMARY (PRIsma for Monitoring AiR quality) research project aims to leverage the capabilities of the Italian Space Agency's PRISMA (PRecursore IperSpettrale della Missione Applicativa) mission to enhance air quality monitoring, particularly in urban areas. The project seeks to utilize PRISMA's hyperspectral data for detailed qualitative and quantitative insights into atmospheric aerosol content and composition in urban environments, crucial for understanding the environmental and health impacts of particulate matter. PRISMA's decametric spatial resolution and the project's use of artificial intelligence address limitations in spatial resolution and the complexity of the inverse problem in satellite-based characterization of particulate matter. In the context of the PRIMARY project, which deals with high-dimensional hyperspectral data, feature extraction before inversion modeling presents challenges, especially when employing machine learning techniques like neural networks (NNs). Dimensionality reduction addresses this challenge by using feature extraction. Comparative evaluation on a PRISMA dataset for Rome showed variable performance between PCA and NN models in compressing and reconstructing the original vector. Additionally, a synthetic dataset was generated to train the algorithm for atmospheric aerosol composition recognition, relying on a sufficiently large number of aerosol profile examples. The Copernicus Atmosphere Monitoring (CAMS) service, specifically its global atmospheric composition forecast product, was chosen as the primary data source. The FlexAOD code, a post-processing tool, was adapted to read CAMS data and obtain aerosol optical properties for input into LibRadtran, a radiative transfer model, used to generate PRISMA-like synthetic data. The resulting dataset, obtained through automated processes, serves as training data for neural networks. To provide validation for the PRIMARY products, the project planned dedicated measurement campaigns in Rome (autumn 2022) and Milan (from winter to summer 2023). The PRIMARY project is co-funded by the Italian Space Agency (ASI Tor Vergata University of Rome Agreement n. 2022-3 U.0); the project is part of the ASIs program PRISMA Scienza.