Distance-based indicators for evaluating environmental multi-contamination and related exposure: how far should we go?

Abstract

Assessing environmental exposure to pollution is a challenging task, and scientists often use distance-based or proximity indicators when field or modeled data are unavailable. Although buffers are commonly used to represent the impact of a pollution source on neighboring populations, they can result in high-exposure misclassification. Euclidean distance-based indicators offer a promising alternative, but practices vary significantly in the literature. In this study, we aimed to compare several distance-based indicators for multiple environmental contaminants in an industrial and urban area. At the population's grid cell resolution of 200 × 200m, we compared the distance to the closest source, the average or median distance to all sources, or a restricted number of nearby sources for six types of sources (industries, railways, rail areas, roadways, road crossings, and agricultural patches) against environmental contamination data (PM10, NO2, and multimetallic contamination in lichens). Our findings revealed that the representativeness of contamination by indicators is significantly affected by the type and number of nearby sources considered. Specifically, we found that considering the distance to the nearest source or the average distance to all sources can lead to exposure misclassifications. The optimal correlation between distance indicators and pollutant levels was observed when considering 10-14 of the closest industrial sources, located within a 4.9- to 5.5-km radius. For rail areas, the optimal number was two to three sources within a 5.4- to 7.4-km radius. For main roads, intersections, and railways, the optimal number of sources varied depending on the pollutant, generally falling within a 3- to 9.4-km radius. Environmental contamination is influenced by the diversity of nearby sources, and considering only one source increases the risk of misclassification. Our results suggest that proximity models are still appropriate for study areas where the etiology of existing health effects is unclear, providing an exploratory analysis before more sophisticated research.