Temporal Variability in an Industrial Building—Time Series and Machine Learning Analysis

Abstract

AbstractInformation is lacking on the degree and causes of temporal variability of indoor air concentrations in industrial buildings. Vapor intrusion (VI) is controlled by multiple variables that interact on different time scales. Indoor concentrations are expected to display multiple periodicities—diurnal, seasonal, and others. An extensive dataset was analyzed using 6‐h time resolution for 6 continuous months including trichloroethylene (TCE), radon, differential pressure, barometric pressure, differential temperature, wind speed, and precipitation. The samples were collected in a sampling zone or compartment within a large industrial building near a point of volatile organic compound (VOC) release. The objectives for this project included assessing VOC temporal variability in an industrial building and evaluating whether using VI indicators/tracers, which are less costly and time intensive to measure, may be able to predict VOC concentrations. This paper reports descriptive, time series, and machine learning data analyses. At this location, radon was proven by far the most effective indicator or tracer for TCE VI. Both the time series and machine learning ranked barometric pressure as a more important influence on VI than temperature. Substantial autocorrelation and diurnal periodicity were observed in TCE concentrations.