SCOUT: Munich Urban Street-level CO 2 Sensor Network

Abstract. The city of Munich has been a pilot city in the ICOS Cities project, which aimed to develop greenhouse gas measurement methodologies and related services for urban areas to support climate action. A part of Munich's city observatory is SCOUT (Street-level Carbon Observatory for Urban Terrain), a network of 100 battery-powered, low-cost, non-dispersive infrared (NDIR) CO2 sensors. This publication describes the sensor network, the applied data processing strategy, and assesses sensor performance. Data from May 2023 to December 2025 are covered. We propose a novel sensor calibration approach that accounts for the strong, non-stationary temperature dependence of the raw sensor signal, thereby improving the manufacturer’s measurement accuracy by a factor of four to eight. In the first stage, a rolling ordinary least squares regression is used to estimate the time-varying, temperature-induced signal component. This is followed by a multilinear regression that infers the final CO2 concentration and corrects for humidity and pressure effects. Evaluated across 21 sensors over three to twelve months outside the training time period at a reference site equipped with a calibrated Picarro analyzer, the proposed model achieves a median RMSE of 9.46 [6.1–12.7] ppm and a R2 of 0.76 [0.55–0.89] for hourly averaged values. It also shows minimal residual dependence on environmental parameters. Long-term drift is corrected by relating the sensor signal to nearby reference stations during favorable atmospheric conditions with homogeneous CO2 concentration across the city. We consider three distinct correction scenarios and estimate a city-wide, location-dependent correction uncertainty of σ between 2.3 and 3.1 ppm. Combined with the measurement uncertainty, this yields a total uncertainty of 9.95 [6.4–13.1] ppm. We deployed the sensor systems in pairs and present a strategy to combine the signals from duplicate measurements at the same site. The network can characterize local emission regimes: winter morning CO2 enhancements show a high correlation with Munich's gridded 2024 emission inventory (r = 0.79), indicating that street-level low-cost networks can serve as proxies for local emissions.