Identification and influence of spatial outliers in air quality measurements

The heterogeneous nature of urban air complicates the analysis of spatial and temporal variability in air quality measurements. Evaluation of potentially inaccurate measurements (i.e., outliers) poses particularly difficult challenges in extensive air quality datasets with multiple measurements distributed in time and space. This study investigated the identification and impact of outliers in measurements of NO­2, BTEX, PM2.5, and PM10 in the contiguous Detroit, Michigan, USA and Windsor, Ontario, Canada international airshed. Measurements were taken at 100 locations during September 2008 and June 2009 and modeled at a 300m by 300m scale resolution. The objective was to determine if outliers were present and, if so, to quantify the magnitude of their impact on modeled spatial pollution distributions. The study built upon previous investigations by the Geospatial Determinants of Health Outcomes Consortium that examined relationships between air pollutant distributions and asthma exacerbations in the Detroit and Windsor airshed. Four independent approaches were initially employed to identify potential outliers: boxplots, variogram clouds, difference maps, and the Local Moran's I statistic. Potential outliers were subsequently reevaluated for consistency among methods and individually assessed to select a final set of outliers. The impact of excluding individual outliers was subsequently determined by revising the spatially variable air pollution models and recalculating associations between air contaminant concentrations and asthma exacerbations in Detroit and Windsor in 2008. For the pollutants examined, revised associations revealed weaker correlations with spatial outliers removed. Nevertheless, the approach employed improves the model integrity by increasing our understanding of the spatial variability of air pollution in the built environment and providing additional insights into the association between acute asthma exacerbations and air pollution.