Abstract. Urban air pollution poses a significant health risk, with over half the global population living in cities where air quality often exceeds World Health Organization (WHO) guidelines. A comprehensive understanding of local pollution levels is essential for addressing this issue. Recent advancements in low-cost sensors and satellite instruments offer cost-efficient complements to reference stations but integrating these diverse data sources in useful monitoring tools is not straightforward. This study presents the updated Retina v2 algorithm, which generates high-resolution urban air pollution maps by assimilating heterogeneous measurements into a portable urban dispersion model. Tested for NO2 concentrations in Madrid during March 2019, it shows improved speed and accuracy over its predecessor, with the ability to incorporate satellite data. Retina v2 balances performance with modest computational demands, delivering similar or better results compared to complex dispersion models and machine learning approaches requiring extensive datasets. Using only TROPOMI satellite data, citywide NO2 simulations show an RMSE of 19.3 µg m−3, with better results when hourly in-situ measurements were included. Relying on data of a single ground station can introduce biases, which can be mitigated by incorporating satellite data or multiple ground stations. Including more stations improves accuracy, with 24 stations yielding a correlation of 0.90 and an RMSE of 13.0 µg m−3. The benefit of TROPOMI diminishes when data from five or more ground stations is available, but it remains valuable for many cities which have limited monitoring networks.
Mijling et al. (Thu,) studied this question.