We built an air quality and public health modeling framework for on-road mobile sources that included emissions inventory development and spatial allocation, meteorological and air quality modeling, combining modeling results and monitored data, and health impact calculations by modifying a prior framework used in an evaluation of heating fuel conversions in buildings [14, 18] (Fig. 1).
Emissions inventory preparation
To characterize baseline conditions we built inventories from EPA’s 2008 National Emissions Inventory (NEI) modeling platform [19]. Described in detail elsewhere, we prepared emissions for three nested grids centered over NYC at 15-km national-scale, 5-km regional-scale, and 1-km local-scale horizontal resolution [18]. We replaced emissions for the on-road mobile source and building heating sectors in the 2008 NEI with more recent, refined local data to better reflect their spatial patterns.
We estimated on-road mobile source emissions using the most recently available county-level data from EPA’s 2011 National Emissions Inventory [20]. County-level emission estimates were spatially allocated to road links in proportion to modeled, link-level vehicle miles traveled from the 2005 New York State Metropolitan Transportation Council (NYMTC) Best Practices Model (BPM) [21]. Despite the relatively older time frame of the NYMTC BPM model, it provided the most recently available modeled counts for cars, trucks, and buses for links within the 28 counties in the NYC region and we assumed that relative spatial patterns in traffic density were reasonably stable between 2005 and 2011. To improve the spatial accuracy within the five NYC counties, the NYMTC shapefile was spatially aligned to the TeleAtlas street segment database within ArcGIS 9.2 Data & Maps.
For grid cells within the 1-km and 5-km modeling grids, we calculated on-road mobile source emissions of total volatile organic compounds (VOC), oxides of nitrogen (NOx), carbon monoxide (CO), sulfur dioxide (SO2), ammonia (NH3), primary PM2.5, and PM2.5 and VOC species profiles. Emissions were allocated do grid cells by first computing at each roadway link the vehicle miles traveled (VMT) by vehicle class (car, truck, bus) by multiplying the annual NYMTC vehicle-specific counts by the length of the segment. We then created ratios, by vehicle type, of the VMTs on each link to the total VMTs in the county. Second, we downloaded the on-road mobile source portion of the 2011 EPA NEI (V1) [20] and matched the Source Classification Code (SCC) subcategories to NYMTC car, truck, and bus categories. All light-duty and heavy-duty gasoline and diesel truck SCC codes were placed in the ‘truck’ category, while light-duty gasoline and diesel vehicles and motorcycles were included in the ‘car’ category. The heavy duty diesel bus SCC codes were included in the ‘bus’ category. Third, we estimated annual link-level emissions for each pollutant and vehicle type by multiplying the county-level emissions by the ratio of the VMTs on each link to the total VMTs in the county. Fourth, we created emissions totals for each pollutant/vehicle type in each 1-km and 5-km grid cell by summing the emissions from links that fell within each grid cell. For links that crossed multiple grid cells, emissions were apportioned based on the fraction of the link’s length included in each grid cell. Finally, because NYMTC does not include VMT breakdowns for categories within ‘car,’ ‘truck’, and ‘bus,’ we approximated these by extracting the county-level VMT data from EPA’s 2008 VMT database [22], then calculated the VMT fractions for gasoline, light- and heavy-duty diesel vehicles. This was then used to assign VOC and PM2.5 speciation profiles by estimating the VOC and PM2.5 emissions for gasoline, light- and heavy-duty diesel vehicles using the VMT fractions for each vehicle type, and then assigning the corresponding PM2.5 and VOC speciation profiles to the each of the categories.
To more accurately represent current building boiler emissions in NYC overall and the within the city, we updated the 2008 NEI for Nos. 2, 4, and 6 heating oil boilers using local permit data reflecting emissions as of 2015. These methods are described in detail elsewhere [14]. Briefly, emissions from Nos. 2, 4, and 6 boilers were calculated using EPA emissions factors [23] and NYC Department of Environmental Protection (NYCDEP) permit data that identify the location and heat throughput of the boiler. No. 4 emissions factors were adjusted to account for NYC-regulated 1500 ppm sulfur content, while No.2 emissions factors assumed a 15 ppm sulfur content, consistent with New York State-wide limits [24]. As many buildings are undergoing conversions of Nos. 4 and 6 boilers to comply with recent regulations [25], we reviewed the permit database as of December 2014, and assigned each building an annual emissions value based on the fuel they were using at that time and spatially allocated these emissions based on boiler address. To estimate emissions from No.2 boilers below the permitting threshold (350,000 Btu), we used NEI emissions not accounted for in the permits, allocating to buildings using surrogate data on building area and the county-specific percent of buildings using No.2 heating oil.
We prepared CMAQ-ready emissions by merging estimates for biogenic sources and all anthropogenic sectors with the updated on-road mobile source inventory and fuel oil boiler inventories. These emissions were processed using the Sparse Matrix Operator Kernel Emissions processor software (version 3.1) to create the air quality modeling input for the base case. We created three additional inventories reflecting removal of specific source categories: zeroing out all motor vehicle emissions within NYC (Sc1), zeroing out truck and bus emissions within NYC (Sc2), and zeroing out all motor vehicles in the 23 counties that surround the five NYC counties (Sc3).
Air quality modeling
Detailed discussion of the application and evaluation of the meteorological and air quality modeling system has been presented elsewhere [18]. In short, meteorological fields for all grids were developed for 2008 using the Weather Research and Forecasting Model (WRF). Air quality modeling was conducted using the Community Multi-Scale Air Quality Model (CMAQ) version 5.0. Annual CMAQ simulations were conducted separately for the base case and each of the three zero-out scenarios and we utilized the daily simulated PM2.5 mass and species concentrations from the 1-km grid cells over NYC for subsequent health burden analyses.
Health burden analysis
Exposure estimates at a 1-km resolution were developed using EPA’s Modeled Attainment Test Software (MATS) [26]. MATS combines the CMAQ modeled output with monitored PM2.5 mass and speciation data to create combined spatial surfaces, providing exposure estimates that use the monitor data but leverages the CMAQ simulated values to better estimate spatial gradients and surface response to changes in emissions. We developed 3 year, quarterly average estimates based on 2010–2012 EPA federal reference monitors (FRM) and speciation trends network (STN) monitors and the daily CMAQ modeling.
We computed the change in number of health events due to changes in PM2.5 concentrations between the base case and each of the three scenarios (Sc1, Sc2, and Sc3) using health impact functions [27, 28]. We applied risk functions for all-cause mortality from chronic exposure among those above 30 years of age [29], emergency department visits for asthma from acute exposure among all age groups (seasonally-specific risk estimates) [30], hospitalizations for all respiratory outcomes from acute exposure among those above 20 years of age (seasonally specific risk estimates for populations above 65 years of age) [31, 32], and hospitalizations for all cardiovascular outcomes from acute exposure among those above 40 years of age (seasonally specific risk estimates) [33]. Risk functions were chosen based on those determined to be most relevant to current New York City populations by selecting those published in peer-reviewed scientific journals and favoring those conducted in New York City when possible. We utilized NYC-specific risk functions for PM2.5-attributable emergency department visits for asthma and hospitalizations for cardiovascular disease. When local studies were not available, we used recent large, multi-city studies or those included in EPA risk analyses [34]. Baseline health data were obtained for 2009–2011 from the NYC Department of Health and Mental Hygiene Bureau of Vital Statistics and the New York Statewide Planning and Research Cooperative System, summarized across 22 age and sex groups within each of 42 United Hospital Fund (UHF) neighborhoods (zip code aggregates). Additional details on risk estimate selection and baseline health data is described elsewhere [27]. Population data for the same age/sex/neighborhood groups were calculated based on the US Census Bureau Population Estimate program [35]. We estimated 3 year, quarterly average health impacts of each of the scenarios within each of the 42 neighborhoods and summed the quarterly estimates to produce annual burdens. All health impact calculations were performed on a quarterly basis using EPA’s Benefits Mapping and Analysis Program (BenMAP) version 4.067 [36]. Further detail on our methodological choices for estimating the public health burden of PM2.5 on NYC residents can be found elsewhere [27].
To estimate years of life expectancy lost (YLL) we calculated life expectancy for 5 year age groupings using the city-wide, baseline mortality rates and standard abridged life table methods from the Centers for Disease Control and Prevention [37]. Years of life lost due to exposures associated with each scenario were calculated by multiplying the number of deaths in each age group attributable to the change in PM2.5 by the remaining life expectancy, then summing across all ages.
We first report the impacts on a citywide basis of removing all traffic in the 28-county region (adding Sc1 heath impacts to Sc3 health impacts), all traffic within NYC (Sc1), trucks and buses within NYC (Sc2), cars within NYC (subtracting Sc2 health impacts from Sc1 health impacts) and traffic from sources within the region but outside of NYC (Sc3). We explore correlations between on-road mobile source category contributions to ambient PM2.5 and neighborhood poverty then grouped neighborhoods based on percent of population residing under the federal poverty threshold (Low: 0–10 %, Medium: 10–20 %, High: 20–30 %, and Very High >30 %), calculated using the 2008–2012 American Community Survey. We report gradients in PM2.5 concentrations, rates of PM2.5-attributable health outcomes, and percent contribution to the total number of health events by neighborhood poverty level.