posted on 2021-02-17, 17:06authored bySheena
E. Martenies, Joshua P. Keller, Sherry WeMott, Grace Kuiper, Zev Ross, William B. Allshouse, John L. Adgate, Anne P. Starling, Dana Dabelea, Sheryl Magzamen
Studies
on health effects of air pollution from local sources require
exposure assessments that capture spatial and temporal trends. To
facilitate intraurban studies in Denver, Colorado, we developed a
spatiotemporal prediction model for black carbon (BC). To inform our
model, we collected more than 700 weekly BC samples using personal
air samplers from 2018 to 2020. The model incorporated spatial and
spatiotemporal predictors and smoothed time trends to generate point-level
weekly predictions of BC concentrations for the years 2009–2020.
Our results indicate that our model reliably predicted weekly BC concentrations
across the region during the year in which we collected data. We achieved
a 10-fold cross-validation R2 of 0.83
and a root-mean-square error of 0.15 μg/m3 for weekly
BC concentrations predicted at our sampling locations. Predicted concentrations
displayed expected temporal trends, with the highest concentrations
predicted during winter months. Thus, our prediction model improves
on typical land use regression models that generally only capture
spatial gradients. However, our model is limited by a lack of long-term
BC monitoring data for full validation of historical predictions.
BC predictions from the weekly spatiotemporal model will be used in
traffic-related air pollution exposure-disease associations more precisely
than previous models for the region have allowed.