State-Informed Background Removal (SIBaR): A Method for Detecting and Removing Background in Mobile Monitoring Campaigns

Abstract:

Mobile air quality monitoring campaigns offer increased spatial coverage at the expense of temporal resolution to better characterize pollution on finer spatial scales. In some mobile monitoring studies, a key step is the quantification and removal of background pollution levels, which allow for the determination of source contributions to mobile measurements. Techniques for background removal vary substantially within the mobile monitoring literature and are often predicated­­­ on static time window assumptions that do not capture the wide variation in source impact temporal scales. Here we discuss a new method for removing background in mobile monitoring data sets, coined SIBaR, that employs Hidden Markov Models (HMMs). We run this method on millions of data points collected in the Houston Mobile Monitoring Campaign, an extensive field program that covers nine months of measurements of nitrogen oxides (NO_x = NO + NO₂) and carbon dioxide (CO₂). The method involves fitting a HMM to log transformed time series data separated by day, taking the points designated as the background state and collecting them, fitting a two-dimensional spline to those points as a function of time and day, and subtracting the resulting signal from the original signal to determine the source contributions. We illustrate proofs of concept of this technique, including mapping the fraction of points designated as background aggregated to each road segment and comparing SIBaR output when it is applied to a second data set pre-identified as background and non-background, finding 86% agreement. We also compare SIBaR-determined source contributions with the source contributions derived using other published time series based background detection techniques. We believe that SIBaR provides a framework for a more consistent background quantification and removal technique.