Measuring Air Quality Levels Inexpensively at Multiple Locations by Random Sampling

Abstract

Collection of a small number of random samples at multiple locations in an urban area offers a means for obtaining air quality data that are spatially more representative—and at the same time less expensive—than fixed-station monitoring at one location. To determine how many samples are needed to attain a desired level of precision, a computer program was written which selects hourly values, at random, from a year of air quality data. The program was run on CO concentrations from an actual city, and results were compared with predictions of the central limit theorem. Predicted and computed results showed excellent agreement, providing a reliable basis for calculating 95% confidence intervals, regardless of whether the original distribution is lognormal or not. To extrapolate these findings to the nation, data from 84 U.S. cities were examined to determine maximum confidence intervals appropriate for different cities. Random sampling was found to give a very precise estimate of the average annual CO concentration with only a small number of samples. With 144 samples, 94% of the U.S. cities had a 95% confidence interval for the annual CO average within ±0.8 ppm. With 100 samples, the 9 5% confidence interval was within ±1 ppm for 94% of the cities. A 75% cost reduction was possible for random sampling compared with fixed-station monitoring approaches.