Local improvement of NO2 concentration maps derived from physicochemical models, using low-cost sensors

Abstract: Urban air quality is a major issue today. Pollutant concentrations, such as NO2's, must be monitored to ensure that they do not exceed dangerous thresholds. Two recent techniques help to map pollutant concentrations on a small scale. First, deterministic physicochemical models take into account the street network and calculate concentration estimates on a grid, providing a map. On the other hand, the advent of new low-cost technologies allows monitoring organizations to densify measurement networks. However, these devices are less reliable than reference devices and need to be corrected. We propose a new approach to improve maps generated using deterministic models by combining measurements from multiple sensor networks. More precisely, we model the bias of deterministic models and estimate it using an MCMC method. Our approach also enables to analyze the behavior of the sensors. The method is applied to the city of Rouen, France, with measurements provided by 4 monitoring stations and 10 low-cost sensors during December 2022. Results show that the method indeed allows to correct the map, reducing estimation errors by about 9.7%.