Stipendiatstilling 1 NILU (2020-2023)

Satellites are incredible tools for studying the atmosphere, offering a global perspective on air quality. However, most satellite instruments measure the total amount of a pollutant across the entire atmosphere, making it difficult to estimate air quality directly at the ground level—where it impacts our health. This PhD research aimed to tackle this challenge by developing advanced algorithms to transform satellite data into accurate surface air quality information. Such improvements are crucial for better assessing human exposure to air pollution, enhancing the societal value of satellite-based air quality data. The research was divided into three interconnected projects, each contributing to a cutting-edge approach for estimating surface-level air pollution. Together, these efforts culminated in the creation of a new model called S-MESH (Satellite and Machine Learning-based Estimation of Surface air quality at High resolution). 1. Creating Surface NO2 Maps: The first project focused on nitrogen dioxide (NO2), a major air pollutant. Using data from the TROPOMI instrument aboard the Sentinel-5P satellite, the research developed a method to estimate NO2 concentrations at ground level. This work provides detailed maps of surface NO2 that are crucial for monitoring urban air quality. 2. Downscaling to 1 km x 1 km Resolution: The second part of the research enhanced the relatively coarse air quality forecasts from the Copernicus Atmosphere Monitoring Service (CAMS), which are available at a resolution of 10 km x 10 km. By incorporating satellite aerosol data and other complementary variables, the study developed a downscaling approach to achieve a much finer resolution of 1 km x 1 km. This improvement makes the data far more relevant for local-scale applications, such as neighborhood-level air quality assessments. 3. Integrating Low-Cost Sensors for City-Level Insights: The final project pushed the boundaries of air quality modeling by incorporating data from low-cost particulate matter (PM) sensors into S-MESH. These sensor networks, which provide dense ground-level measurements, were seamlessly integrated into the model to further refine its predictions. This not only enhanced model accuracy but also opened the door to even higher spatial resolution predictions, potentially enabling within-city assessments of air quality. Such advancements are critical for understanding fine-scale pollution patterns and addressing urban air quality challenges. At the heart of this thesis was the innovative use of machine learning, particularly a method called XGBoost, to handle the complexity of combining multiple data sources and predictor variables. This approach highlights the power of machine learning to tackle real-world challenges in environmental science, providing tools to better understand and address air pollution.

The project has delivered significant impacts for participants, stakeholders, and society, particularly in advancing research, environmental monitoring, and policymaking. By developing the S-MESH framework (Satellite and Machine Learning-based Estimation of Surface air quality at High resolution), the project has strengthened competence in satellite data utilization, machine learning, and environmental modeling. This innovative approach has already been integrated into ongoing projects and new proposals. The knowledge and methodologies developed have enhanced NILU’s research capacity, fostering a deeper integration of machine learning and remote sensing expertise with atmospheric science. The project has also promoted interdisciplinary and international collaboration, expanding partnerships with leading institutions in satellite data analysis and environmental monitoring. This collaborative environment has not only strengthened research outputs but also laid the foundation for further projects, including new proposals to Horizon Europe addressing Europe’s zero-pollution goals. By producing high-resolution air quality maps, the project has developed tools with immediate practical relevance for urban planners, policymakers, and environmental agencies. These tools can help identify pollution hotspots, improve urban air quality management, and align with the EU’s Green Deal objectives. Looking forward, the long-term effects of the project are expected to be substantial. By delivering accurate and timely information on surface-level air pollution, the project contributes to improved public health outcomes. Reducing human exposure to pollutants can mitigate respiratory illnesses and reduce premature mortality linked to air quality issues. Moreover, the S-MESH framework represents a technological innovation that showcases how machine learning and satellite data can be effectively combined to address pressing environmental challenges. This model has the potential to be further developed and the underlying methods can in principle be applied globally. The project’s outcomes also align with broader climate and environmental goals. By providing tools to monitor and reduce air pollution, it supports international efforts such as Europe’s zero-pollution targets and global environmental monitoring initiatives under the United Nations frameworks. Through these contributions, the project not only advances scientific knowledge but also fosters sustainable urban development as well as progress in satellite-based environmental monitoring.