Abstract View
Openly Accessible Low-Cost Measurements for PM2.5 Exposure Modeling: Guidance for Monitor Deployment with a Similarity Metric
JIANZHAO BI, Nancy Carmona, Magali Blanco, Amanda Gassett, Edmund Seto, Adam Szpiro, Timothy Larson, Paul Sampson, Joel Kaufman, Lianne Sheppard, University of Washington
Abstract Number: 377
Working Group: Aerosol Exposure
Abstract
High-resolution, high-quality exposure modeling is critical for assessing the health effects of PM2.5 in epidemiological cohorts. Sparse ground-level PM2.5 measurements, as key model input, may result in two potential issues in high-resolution exposure prediction: (1) they may affect the models’ accuracy in predicting the spatial distribution of PM2.5; and (2) internal validation based on these measurements may not reliably reflect the model performance at the locations of interest (e.g., ambient PM2.5 concentrations at cohort residential locations). This study aimed to use PM2.5 measurements from an openly accessible low-cost PM2.5 network, PurpleAir, with an external validation dataset at residential locations of a representative sample of participants from the Adult Changes in Thought - Air Pollution (ACT-AP) study to improve the accuracy of exposure prediction at the cohort locations, and propose a metric assessing the similarity between the monitor and cohort locations to guide future monitor deployment. We utilized a spatiotemporal modeling framework to incorporate PM2.5 measurements from 51 “gold-standard” monitors and 58 PurpleAir monitors in the Puget Sound region of Washington into high-resolution exposure assessment at the two-week level from June 2017 to March 2019. We proposed a similarity metric based on principal component analysis (PCA) - the PCA distance - to assess the PurpleAir monitors’ representativeness of the cohort locations. After including calibrated PurpleAir measurements as part of the dependent variable, the spatiotemporal validation (at the two-week level) R2 and root-mean-square error, RMSE, improved from 0.84 and 2.22 μg/m3 to 0.92 and 1.63 μg/m3. The spatial validation R2 and RMSE improved from 0.72 and 1.01 μg/m3 to 0.79 and 0.88 μg/m3. We found that the PurpleAir monitors with shorter PCA distances could improve the model’s prediction accuracy more substantially than monitors with longer PCA distances, indicating the reasonability of this similarity metric. To our knowledge, this was the first attempt to evaluate the benefits of low-cost PM2.5 measurements for long- and short-term exposure prediction at cohort residential locations and to provide practical guidance with a PCA-based similarity metric for future monitor deployment.