Spatial and temporal approaches for differentiating local source and background air quality impacts using low-cost sensor networks

Abstract. Low-cost sensor (LCS) networks are increasingly used to characterize the impacts of local air pollution sources. Researchers have developed a variety of methods to differentiate local impacts from background influences; however, these methods differ substantially in their definitions, assumptions, and resulting insights.  This study compares four spatial and temporal techniques for isolating local influences: Spatial Network, Peak Finding, Fourier Decomposition, and Wavelet Decomposition. We apply these techniques to three pollutants across two LCS networks to assess their strengths, limitations, and suitability for different contexts. The Spatial Network approach performed better for pollutants with spatially localized plumes and low regional background, such as coarse particulate matter, but was sensitive to systematic sensor bias. In contrast, the temporal approaches were more robust to sensor bias but were instead limited by their underlying assumption of the short-lived nature of local impacts. The Peak Finding approach was effective for identifying episodic plume events but struggled to differentiate local impacts for pollutants with strong diurnal variability, such as nitrogen dioxide. Wavelet Decomposition and Fourier Decomposition provided greater flexibility for separating variability across multiple temporal scales, such as diurnal, weekly, and seasonal cycles. Each method also showed distinct tradeoffs in implementation complexity, parameter sensitivity, and data completeness requirements. These findings demonstrate that researchers should select a technique according to the community context, pollutant characteristics, and data quality rather than applying a single approach universally.