Improving Urban Eddy-Covariance CO2 Flux Estimates Through Removal of Anomalies in High-Frequency Data Using the IMAS Algorithm

Urban areas are major sources of anthropogenic CO₂ emissions, contributing substantially to the global carbon budget. Accurate quantification of urban emissions remains challenging due to uncertainties in measurements and modelling approaches. Eddy covariance (EC) provides direct continuous measurements of net urban CO₂ fluxes; however, flux estimates in heterogeneous urban environments can be systematically biased by unresolved micro-scale anthropogenic sources. This study investigates the efficiency of the Identification of Micro-scale Anthropogenic Sources (IMAS) algorithm (Kotthaus and Grimmond, 2012) to detect short-duration, high-frequency micro-scale signals on EC observations. IMAS removes statistically identified micro-scale events from high-frequency data prior to flux computation, enabling retention of standard 30-min averaging periods. Micro-scale event detection is based on statistical metrics computed at 1-min resolution for CO₂, H₂O and sonic temperature, combining kurtosis, median-based variability, and skewness-sensitive mid-range deviation referenced to a 30-min median.

We applied the IMAS algorithm to two years of continuous EC measurement data, which were collected at the Hardau tall-tower site in the city of Zurich, Switzerland, as part of the ICOS Cities project. Fluxes were measured on a mast on top of a high-rise building at 112 m a.g.l, sampling a heterogeneous footprint influenced by various sources such as residential heating, traffic, railway infrastructure and industrial activities. A local heating unit is located at a horizontal distance of 145 m south-east of the tower, which is used intermittently to support residential heating during cold periods and could potentially affect our tower measurements. Standard EC fluxes and quality control flags were computed using EddyPro software before (L1) and after (L2) the application of the IMAS algorithm. Flux differences between L1 and L2 show a strong dependence on wind direction, with the largest reductions in L2 occuring for sector spanning 120–160°, centered on the direction of a nearby local heating unit (~141°) within the urban footprint. During winter, standard EC processing (L1) overestimates CO₂ fluxes by 3.96 ± 0.43 µmol m^-2 s^-1 (mean ± standard error of the mean) for wind originating from this sector, corresponding to a relative reduction of ~17 % after the IMAS-based removal of micro-scale events. Smaller but consistent mean reductions are also observed for H₂O fluxes (0.039 ± 0.005 mmol m^-2 s^-1, ~12 %) and sensible heat fluxes (4.82 ± 0.75 W m^-2, ~38 %). In contrast, IMAS-induced flux changes during summer were minimal. These results demonstrate that unresolved micro-scale emissions can propagate directly into urban CO₂ flux calculations, highlighting the need for source-aware, high-frequency preprocessing to complement standard EC quality control in urban carbon flux monitoring.