Other team members:
 

Scientific Focus of Data Sets: Earth systems science

Description of Data Sets:

This study uses the following air pollutant concentration datasets, all of which Veronica Tinney Southerland, under the supervision of Dr. Susan Anenberg at the George Washington University, have permission by the dataset authors for use in this study.
1. Di et al. 2016 satellite derived fine particulate matter dataset. This dataset, created by Di and colleagues, is a hybrid model that produced daily estimates of fine particulate matter for the year 2016. The neural network trained hybrid model incorporates aerosol optical depth from the Moderate Resolution Imaging Spectroradiometer (MODIS) on the Earth Observing System (EOS) satellite, as well as monitoring data, a chemical transport model, as well as meteorological and land use variables.
2. Google Street view mobile monitoring concentrations. The Environmental Defense Fund (EDF) and Google Earth partnered to measure nitrogen dioxide through Google Street View cars in Oakland, California. The results are available from Apte et al. 2017. The authors outfitted two Google Street View cars with rapid air pollution monitors for nitrogen dioxide. Between May 2015 and May 2016, the cars were given drive areas which resulted in repeated measurements for every 30 meters in Oakland5, resulting in over three million 30 meter data points. The authors then computed the mean for each 30 meter segment to create a year-long study period average.
3. Larkin et al. 2017 global land use regression dataset for nitrogen dioxide. This dataset, created by Larkin and colleagues, provides global estimates of nitrogen dioxide at 100 meter resolution, for the year 2011.

Scientific Potential of Presentation:

Estimates of the health impacts associated with ambient air pollution in the United States are typically reported at the state or county level, which obscures potential heterogeneity in impacts at finer spatial scales. While informative, the spatial distribution of health impacts can reveal particular locations that may be experiencing greater than average impacts, as well as inequality in which neighborhoods and population sub-groups are most affected. Estimating air pollution health impacts at the hyper-local scale (i.e. 100m x 100m) is now possible with concentrations derived from satellite remote sensing and mobile monitoring. In 2015, the Environmental Defense Fund and Google Earth conducted mobile monitoring of black carbon and nitrogen dioxide (NO2) in the Bay Area using Google Street View cars outfitted with a pollution measurement platform. This study estimates the health impacts at 100m resolution throughout the Bay Area using satellite-derived fine particulate matter (PM2.5) estimates, NO2 measurements from mobile monitoring, and NO2 estimates from a land use regression model. This study explores how estimated health impacts differ when using mobile monitoring versus satellite and land use regression concentration estimates, and when using higher resolution versus county level baseline disease rates. Initial results demonstrate that the aggregated air pollution-attributable disease burdens for the Bay area are similar across models using varying inputs. However, using highly resolved baseline disease rates and the mobile monitoring concentration datasets captures spatial variability that is obscured when more coarsely resolved data inputs are used.

Data visualization is particularly important for this study as initial results show that the improved ascertainment of spatial distribution of results may be more impactful for local decision makers in understanding how air pollution affects their locality as well as where to target interventions to reduce overall air pollution health impacts and mitigate health inequalities. For example, use of county level baseline disease rates as compared to census block group rates results in census block group aggregated estimates that are over-estimated in some census blocks, and underestimated in others, as compared to the results received from models incorporating census block group disease rates. Without use of highly resolved baseline disease rates, decision-makers may come to different conclusions about where to best target interventions or identify disparities in risk. Visualizing the results of the risk estimates, rather than aggregating them to the county or Bay area level, allows for the identification of spatial patterns in risk that may not have previously been apparent in courser resolution risk estimates. For example, initial results so that use of census block disease rates results in census block groups that are on average 25-50% higher than those received using county baseline disease rates. It is the hope of this study that the results will not only inform best practices for conducting health impact assessment on the hyperlocal scale, but also demonstrate the importance in visualizing spatial patterns of risk.