A Statistical Assessment of Saturation and Mobile Sampling Strategies to Estimate Long-Term Average Concentrations across Urban Areast

Abstract

The objectives of this study were: (1) to quantify the errors associated with saturation air quality monitoring in estimating the long-term (i.e., annual and 5 yr) mean at a given site from four 2-week measurements, once per season; and (2) to develop a sampling strategy to guide the deployment of mobile air quality facilities for characterizing intraurban gradients of air pollutants, that is, to determine how often a given location should be visited to obtain relatively accurate estimates of the mean air pollutant concentrations. Computer simulations were conducted by randomly sampling ambient monitoring data collected in six Canadian cities at a variety of settings (e.g., population-based sites, near-roadway sites). The 5-yr (1998–2002) dataset consisted of hourly measurements of nitric oxide (NO), nitrogen dioxide (NO2), oxides of nitrogen (NOx), sulfur dioxide (SO2), coarse particulate matter (PM10), fine particulate matter (PM2.5), and CO. The strategy of randomly selecting one 2-week measurement per season to determine the annual or long-term average concentration yields estimates within 30% of the true value 95% of the time for NO2, PM10 and NOx. Larger errors, up to 50%, are expected for NO, SO2, PM2.5, and CO. Combining concentrations from 85 random 1-hr visits per season provides annual and 5-yr average estimates within 30% of the true value with good confidence. Overall, the magnitude of error in the estimates was strongly correlated with the variability of the pollutant. A better estimation can be expected for pollutants known to be less temporally variable and/or over geographic areas where concentrations are less variable. By using multiple sites located in different settings, the relationships determined for estimation error versus number of measurement periods used to determine long-term average are expected to realistically portray the true distribution. Thus, the results should be a good indication of the potential errors one could expect in a variety of different cities, particularly in more northern latitudes.