Design, operation, and insights from Zürich city's mid- and low-cost CO2 sensor network

As a part of the ICOS Cities project, a dense CO₂ sensor network was deployed across Zürich city in July 2022 that will remain operational until July 2024. The network comprises 250 NDIR (nondispersive infrared) CO₂ sensors from three manufacturers (Senseair, Vaisala, and Licor) at 87 monitoring sites. The sensors can be classified into low- and mid-cost groups (~€500 and ~€7000 respectively). Most mid-cost sensors were installed with rooftop inlets, while most low-cost sensors were deployed near ground level, i.e. near sources of biogenic activities, human respiration, and fossil fuel burning. All data are transferred using LoRaWan and Picarro CRDS (cavity ring-down spectroscopy) gas analysers with traceable reference gases are used for calibration and the assessment of the sensors’ performance before field deployment.
The mid-cost CO₂ sensors run on mains power, are placed inside maintenance rooms or measurement cabins, and make use of two calibration gases that are tested daily. After accounting for air pressure, humidity and the reference gas tests, the mid-cost CO₂ sensors achieve an accuracy of 1.5 ppm of root mean square error (RMSE) and a mean bias that is within ± 1 ppm when considering hourly means in field conditions. The low-cost sensors are battery-powered and require an initial calibration period to address potential deficiencies with their factory calibration. During field deployment, an algorithm for drift correction is applied that considers meteorological conditions and data provided by the mid-cost sensors in the network. The low-cost sensors achieve a mean RMSE of 15 ppm under field conditions when compared to pseudo-reference time series provided by mid-cost sensors, and on average, they show no systematic bias.
The sensor measurement performance is adequate to resolve site-specific differences and interesting source-sink processes – especially those related to traffic and the biosphere. Mean CO₂ dry air mole fractions ranged between 432 and 460 ppm across the network with some sites displaying large CO₂ diurnal ranges (up to 70 ppm) due to confinement of biogenic emissions in the very early hours of the morning. The network’s background CO₂ is highly variable, indicating that Zürich’s ambient CO₂ levels are strongly influenced by regional scale processes as well as emissions and sinks within the city’s boundary. In a first attempt to quantify CO₂ emissions, the rooftop sensors are combined with inventory data and simulations of biogenic activity using ICON-ART at a spatial resolution of 600 m. In contrast, the low-cost sensors will be employed in combination with highly-resolved urban emission data and GRAMM/GRAL, a building-resolved transport model. In collaboration with the city government, we expect this to become a long-term, actionable contribution to address urban emissions and the city's net-zero commitment.