Automated identification of local contamination in remote atmospheric composition time series

Abstract. Atmospheric observations in remote locations offer a possibility of exploring trace gas and particle concentrations pristine environments. However, data from areas are often contaminated by pollution local sources. Detecting this contamination is thus central frequently encountered issue. Consequently, many different methods exist today to identify atmospheric composition measurement time series, but no single method has been widely accepted. In study, we present new primary datasets, e.g., ship campaigns or stations with low background signal compared the signal. The detection algorithm (PDA) identifies flags periods polluted five steps. first most important step based on derivative (time derivative) concentration over time. If exceeds given threshold, flagged as polluted. Further identification steps simple threshold filter, neighboring points filter (optional), median, sparse (optional). PDA only relies target dataset itself independent ancillary datasets such meteorological variables. All parameters each adjustable so that can be “tuned” more less stringent (e.g., flag fewer contaminated). was developed tested number collected during Multidisciplinary drifting Observatory for Study Arctic Climate (MOSAiC) expedition Arctic. Using strict settings, identified 62 % influenced contamination. second also MOSAiC, evaluated performance against same cleaned visual inspection. two agreed 94 cases. Additionally, successfully applied (CO2), another dataset, at high-altitude station Jungfraujoch, Switzerland. Thus, proves useful flexible tool affected without need measurements. It best representing pollution. user-friendly open-access code enables reproducible application wide suite datasets. available https://doi.org/10.5281/zenodo.5761101 (Beck et al., 2021).