Citizen science projects that monitor air quality have recently drastically expanded in scale. Projects involving thousands of citizens generate spatially dense data sets using low-cost passive samplers for nitrogen dioxide (NO2), which complement data from the sparse reference network operated by environmental agencies. However, there is a critical bottleneck in using these citizen-derived data sets for air-quality policy. The monitoring effort typically lasts only a few weeks, while long-term air-quality guidelines are based on annual-averaged concentrations that are not affected by seasonal fluctuations in air quality. Here, we describe a statistical model approach to reliably transform passive sampler NO2 data from multiweek averages to annual-averaged values. The predictive model is trained with data from reference stations that are limited in number but provide full temporal coverage and is subsequently applied to the one-off data set recorded by the spatially extensive network of passive samplers. We verify the assumptions underlying the model procedure and demonstrate that model uncertainty complies with the EU-quality objectives for air-quality monitoring. Our approach allows a considerable cost optimization of passive sampler campaigns and removes a critical bottleneck for citizen-derived data to be used for compliance checking and air-quality policy use.