(ISCRA project Class C) – Continuous super-resolution of primary air pollutants using implicit diffusion models

Organization

Program

Executive Summary:

Air pollution remains a persistent environmental and public health challenge in developed Western countries, with primary pollutants such as fine particulate matter, nitrogen dioxide, ozone, and sulphur dioxide posing serious risks. Accurate, high-resolution data on ground-level concentrations of these pollutants is essential for effective environmental monitoring, health impact assessments, and the development of informed public policy. However, existing air quality monitoring solutions often lack the spatial detail needed, due to limitations in sensor coverage, resolution, and cost. This project addresses these challenges by enhancing the spatial resolution of large-scale atmospheric chemistry datasets, specifically the CAMS Ensemble products and CAMS Global Reanalysis, using Implicit Diffusion Models. The proposed model, integrated with Graph Attention Network (GAT), performs a continuous super-resolution technique to generate fine-scale pollutant concentration maps that better reflect localized variations, particularly in urban areas and complex terrains. A key innovation of this approach is its ability to provide Bayesian uncertainty quantification, allowing for robust, interpretable predictions that support confident decision-making by governments and public health authorities. By quantifying the reliability of the model’s outcomes, the project ensures that pollutant redictions are not only high-resolution but also reliable. Additionally, the use of GATs enables the model to learn spatial dependencies more effectively by dynamically focusing on relevant spatial relationships among data points. To ensure the accuracy and trustworthiness of the model’s probabilistic outputs, calibration tests will be performed, with re-calibration procedures applied if necessary. The project aims to enhance the spatial resolution of ground-level primary air pollutant data, derived from the CAMS Ensemble and CAMS Global Reanalysis atmospheric chemistry datasets, across Italy, covering the period from 2013 to 2023.