Detailed validation of air quality models is essential, but remains challenging, due to a lack of suitable high-resolution measurement datasets. This is particularly true for pollutants with short-scale spatial variations, such as nitrogen dioxide (NO2). While street-level air quality model chains can predict concentration gradients at high spatial resolution, measurement campaigns lack the coverage and spatial density required to validate these gradients. Citizen science offers a tool to collect large-scale datasets, but it remains unclear to what extent such data can truly increase model performance. Here we use the passive sampler dataset collected within the large-scale citizen science campaign CurieuzeNeuzen to assess the integrated ATMO-Street street-level air quality model chain. The extensiveness of the dataset (20.000 sampling locations across the densely populated region Flanders, ∼1.5 data points per km2) allowed an in-depth model validation and optimization. We illustrate generic techniques and methods to assess and improve street-level air quality models, and show that considerable model improvement can be achieved, in particular with respect to the correct representation of the small-scale spatial variability of the NO2-concentrations. After model optimization, the model skill of the ATMO-Street chain significantly increased, passing the FAIRMODE model quality threshold, and thus substantiating its suitability for policy support. More generally, our results reveal how a “deep validation” based on extensive spatial data can substantially improve model performance, thus demonstrating how air quality modelling can benefit from one-off large-scale monitoring campaigns.@en