Green Heart Louisville: intra-urban, hyperlocal land-use regression modeling of nitrogen oxides and ozone

Exposure to urban air pollution is linked to increased mortality from cardiopulmonary causes. Urban areas juxtapose large numbers of residences and workplaces with near-road environments, exacerbating traffic-related air pollution (TRAP) exposure. TRAP is the primary source of variability in intraurban air quality, but continuous regulatory monitoring stations lack the spatial resolution to detect fine-scale pollutant patterns that recent studies using long-term, resource-intensive mobile measurements have established as persistent and associated with higher risk of cardiovascular events. This work evaluates a low-cost, fixed-site approach to characterizinglong-term, hyperlocal exposure to oxides of nitrogen (including NO2, a common surrogate for TRAP) as part of Green Heart Louisville, a prospective cohort study examining linkages between urban vegetation, local air quality, and cardiovascular health. We used a fixed 60-site network of Ogawa passive samplers in a 12 km2 section of Louisville, KY, to measure two-week integrated NO2, NOx (NO + NO2), and O3 mixing ratios nominally every two months between May 2018-March 2021. Seasonal NOx averages were 2.5-fold higher during winter than in summer, and annual average NO (calculated by difference in NOx and NO2) and NO2 ranged from 4-21 ppb and 5-12 ppb, respectively. NO increased 3-to-5-fold within 150 m of highways or major arterial roads and 2-to-3-fold near parking lots. While both NO and NO2 were elevated in near-road environments, the corresponding O3 was depressed, consistent with titration by NO. We developed land-use regression models for annual average NO, NO2, and NOx using parameters of proximity (distance to nearest road type, restaurant, traffic signal), cumulative occurrence (length of roads, number of restaurants and traffic lights, all in buffers of up to 500 m in 50-m increments), and greenness (normalized difference vegetative index (NDVI)). Adjusted spatial variability explained by the models were 70% (p<0.05), 67% (p<0.05), and 75% (p<0.01) for NO, NO2, and NOx, respectively. Common predictors were distances to the nearest restaurant and road as well as total length of roads within 350 m. Only one greenness metric was significant: mean NDVI within 50 m was negatively associated (p=0.02) with NO2. We plan to use these hyperlocal models to estimate residential-level exposures of the clinical study participants.