10th International Aerosol Conference September 2 - September 7, 2018 America's Center Convention Complex St. Louis, Missouri, USA
Abstract View
Data Requirements for Mapping Long-Term Air Pollution with Mobile Short-Term Measurements
RIVKAH GARDNER-FROLICK, Joshua Apte, Kyle Messier, University of Texas at Austin
Abstract Number: 1093 Working Group: Aerosol Exposure
Abstract New large-scale approaches to air quality data collection have the potential to substantially advance understanding of human exposure to outdoor air pollution, which varies sharply in space and time. For example, some methods use mobile monitors or temporarily stationed sensors to provide highly spatially resolved data. One inherent limitation is that these methods provide sparse temporal resolution for any given location, raising the question of their ability to monitor long-term trends. Here, we investigate whether a series of short-term measurements can adequately represent long-term average concentrations at a monitoring site. To do so, we conduct a Monte Carlo subsampling analysis of 2011-2017 Bay Area Air Quality Management District (BAAQMD) 1-hour measurements of black carbon (BC), nitric oxide (NO), nitrogen dioxide (NO2), fine particulate matter (PM2.5), carbon monoxide (CO), sulfur dioxide (SO2), and ultrafine particulate matter (UFP) at multiple locations in the San Francisco Bay Area, California and 2012-2014 Southeastern Aerosol Research and Characterization (SEARCH) Network 1-minute measurements of BC, NO, NO2, and CO in Atlanta, Georgia. Methods developed using the BAAQMD and SEARCH data ultimately will be applied to air pollution data collected in the San Francisco Bay Area using Google Street View cars. In this dataset, 1-Hz measurements of air pollution are matched to the nearest 30-meter road point and analyses can be performed for each location or for larger road segments.
We use random and structured subsampling to estimate the number of observations needed to achieve an accurate (within ±10%) estimation of the annual mean. The geometric standard deviation of continuous measurements, for both 1-minute and 1-hour time intervals, is descriptive of air quality at a site and is highly predictive of the number of samples required to estimate the mean for a specific pollutant at that site. Across all locations, pollutants tend to cluster together, having a similar geometric standard deviation and required number of samples. The pollutant with the lowest number of required samples is CO, which requires 32 independent observations. The pollutant with the highest required number is NO, with 428 samples. While pollutants tend to cluster together across all monitored locations, a few points have significantly different required sample numbers. For example, the average required sample number for BC is 112, while the required sample number at the Forest Knolls site is 490. These breaks in clustering can be attributed to extremely variable seasonal and diurnal patterns, with the breaks occurring at a given location in a predictable manner. These overall trends hold for both random and structured sampling schemes, corresponding to different real-world field study designs. Given the highly correlated nature of air quality measurements, we found that independent measurements are more important than the length of the measurement period. The results hold important implications for future air quality study design and the practicality of using disparate monitoring methods to collect the most useful information. A small number of discrete measurements, such as those gathered from mobile monitoring, can be highly predictive of annual averages.