Modeling nitrogen dioxide concentrations using citizen science data: The case of the Brussels-Capital Region

Abstract

Air pollution caused by NO${}_2$ emissions related to traffic is a major environmental issue in the Brussels-Capital region. Using a large set of measurements collected from a citizen science campaign, this paper shows how such data help us to get an overview of the spatial distribution of NO${}_2$ levels over the region. Using two land use regression techniques, these levels were related to spatial proxies collected at the measurement locations. Comparing the proxies selected by each regression method offers deeper insights into the NO${}_2$-proxies relationships and helps identify proxies that may have been overlooked in a simpler multilinear regression model. Results show that the multiple linear regression model is able to explain a major part of the variance of the data, while random forest regression performs slightly better, with performances that are on par with those found in the literature. However, both models tend to underestimate high concentrations that are occurring locally. Thanks to a comparison with the prediction results from a physics-based model, this could be related to the quality of the input traffic data, that are expected to play a major role as most of nitrogen oxides emissions in the Brussels-Capital region originate from road traffic.