Authors: Ekaterina Galkina, Georges Grinstein
Abstract: Growing evidence points to serious medical conditions associated with airborne pollutant exposure, including neurologically harmful effects, increases in hospitalizations and mortality due to cardiovascular and respiratory diseases. The US Environmental Protection Agency (EPA) regularly monitors levels of lead, ozone, particulate and other air pollution. However, monitor sites are unequally distributed, necessitating spatial interpolation of pollutant concentrations at poorly sampled locations. Our study focuses on epidemiological data aggregated to only the first 3-digits of the postal ZIP code. Consequently, the varying land area coverage of the information-sensitive residential geographies creates an additional exposure assessment challenge. We compared four common interpolation methods for predicting particulate matter concentrations in the state of California alone, including kernel smoothing, inverse distance weighting (IDW), Voronoi partitioning and kriging. We show that methods that produce a prediction standard error map are the most reliable (i.e. kriging) and that a consensus in the prediction maps can be reached only at the smaller geographical units, i.e. county-level.