Use of Population Exposure Frequency Distributionsto simulate effects of policy interventions on NO2exposure

C. DIMITROULOPOULOU1, M.R.ASHMORE2*, A.C TERRY2

1: Environmental Change Department, Centre for Radiation, Chemical and Environmental Hazards, Public Health England,Chilton, OX11 0RQ

2. Stockholm Environment Institute, Environment Building, Wentworth Way, University of York, York YO10 5NG, UK

*corresponding author: e-mail:;

tel: 00-44-1904-324744

Abstract

Health effects of air pollution on individuals depend on their personal exposure, but few modelling tools are available which can predict how the distribution of personal exposures within a city will change in response to policies to reduce emissions both indoors and outdoors. We describe a new probabilistic modelling framework (INDAIR-2/EXPAIR),which provides predictions of the personal exposure frequency distribution (PEFD) across a city to assess the effects of both reduced emissions from home sources and reduced roadside concentrations on population exposure. The model uses a national time activity database, which gives the percentage of each population group in different residential and non-residential micro-environments, and links this, for the home, to predictions of concentrations from a three-compartment model, and for non-residential microenvironments to empirical indoor/outdoor ratios. This paper presents modelled PEFDs for NO2in the city of Leicester, for children, the elderly, and office workers, comparing results in different seasons and on different days of the week. While the mean NO2 population exposure was close to, or below the urban background concentration, the 95%ile of the PEFD was well above the urban background concentration. The relationship between both mean and 95%ile PEFD and urban background concentrations was strongly influenced by air exchange rate. The 24h mean PEFD showed relative small differences between the population groups, with both removal of home sources and reductions of roadside concentrations on roads with a high traffic density having similar effects in reducing mean exposure. In contrast, the 1h maximum of the PEFD was significantly higher for children and the elderly than for office workers, and showed a much greater response to reduced home emissions in these groups. The results demonstrate the importance of understanding the dynamics of NO2 exposure at a population levelwithin different groups, if the benefits of policy interventions are to be accurately assessed.

Key words: personal exposure, indoor air pollution, modelling; nitrogen dioxide

Highlights

  • A new model predicts personal exposure frequency distributions (PEFD) across a city
  • It links a microenvironmental model of the home to time-activity patterns
  • We applied the model to simulate PEFDs for nitrogen dioxide in the city of Leicester
  • The mean of the PEFD was similar to, or below, urban background concentrations
  • The 95%ile PEFD was more strongly influenced by indoor and outdoor emission peaks

1. Introduction

Over the past two decades, the short-term and long-term adverse effects on human health of exposure to relatively small concentrations of particulates have been established and quantified (e.g. COMEAP 2010; Atkinson et al., 2014a;b), but there is now increasing evidence of the effects of nitrogen dioxide (NO2), both in the home (e.g. Logue et al., 2011) and outside (e.g. Faustini et al., 2014;Mills et al., 2015).Indoor and outdoor exposure to NO2 can affect the health of childrenand the elderly in particular(e.g. Breysse et al., 2010;Favarato et al., 2014; USEPA, 2016; RCP, 2016). A recent meta-analysis (Lin et al., 2013) confirmed that both gas cooking and elevated indoor NO2 concentrations were associated with a higher prevalence of childhood asthma and wheeze; furthermore, there is evidence that the threshold for increased symptoms in asthmatic children, as a result of long-term exposure, lies below 10ppb i.e. below current annual air quality guidelines (Belanger et al., 2013). In adults, increased bronchial responsiveness has been associated with gas cooking for individuals without a gene which increases oxidant stress detoxification (Amaral et al., 2014).

The evidence associating NO2exposure with health effects has strengthened substantially in recent years (COMEAP, 2015).Three authoritative reviews on the health effects from exposure to NO2 have been carried out: a statement on the quantification of the effects of long-term exposure to NO2 on respiratory morbidity in children by COMEAP (2009), the updated Integrated Science Assessment by US EPA (2013, 2016), and the Review of Evidence on Health Aspects of Air Pollution-REVIHAAP by WHO (2013). In addition, recent studies, including the ESCAPE study (Beelen et al., 2014), the DUELS study (Fischer et al., 2015),a study following the RoLS cohort in Rome (Cesaroni et al., 2013)and a meta-analysis of long-term studies on NO2 (Faustini et al, 2014), show associations between long-term exposure to NO2and all-cause, respiratory and cardiovascular mortality, children’s respiratory symptoms and lung function. Furthermore, there are positive associations between short-term exposure to NO2 and hospital admissions and emergency room visits for cardiovascular and/or cardiac diagnoses. Positive and statistically significant associations of short-term ambient concentrations of NO2 with all-cause and cause-specific mortality have also been reported. Although the evidence for short-term effects is stronger, whether these associations reflect adverse health effects of long-term exposure to NO2 or other pollutants emitted by thesame sources, such as traffic, has been a matter of debate(COMEAP, 2015).

Although some epidemiological studies of the effects of indoor NO2exposure use direct measurements in the home, most outdoor studies have either assumed that exposure of the study populationscan be represented by measurements made at fixed central monitoring sites, or have used appropriate predictive models to estimate the outdoor concentrations at the place of residence. However, health outcomes are most directly related to the personal exposure of individuals,and there is evidence that day-to-day variation in symptoms is more closely related to measured personal exposure than to data from fixed-site monitors (e.g. Spira-Cohen et al., 2011; Brook et al., 2011).Several studies of NO2 exposure assessment have examined factors that affect thereliability of NO2 ambient monitors to act as a surrogatefor personal NO2 exposures in epidemiological studies, given that thestrength of the association between ambient NO2 and personal NO2 exposure contributes to exposure error (e.g. Vardoulakis et al., 2011; Brown et al., 2009; Delfino et al., 2008; Kim et al., 2006). In some cases,data from fixed monitoring stations explain only a fraction of variation in personal NO2 exposures, which are more strongly associated with the NO2 indoor residential and workplace concentrations (e.g. Kousa et al., 2001; Alm et al., 1998).Meng et al. (2012), in a meta-analysis of studies carried out over the previous 30 years, concluded that the strength of the associations between personal NO2 exposure and outdoor concentrations, although positive overall, varied among studies, with differences related to study design and exposure factors.

Furthermore, for a given outdoor pollutant concentration field across a city, there will be a wide range of individual personal exposures within the population, due to the effects of exposure in different indoor micro-environments, and variations in people’s activity, added to spatialvariation in outdoor concentrations.Personal exposures can be determined directly by measurement, or indirectly using computer models. When the aim is to assess the effect of different policy interventions on population exposure, models offer significant advantages over direct measurement methods, as they allow the relative contributions of different indoor and outdoor sources to population exposure to be assessed, and the benefits of different policy measures to be compared (Dimitroulopoulou et al., 2001). Furthermore, when the distribution of exposures within a population is of interest, a very large number of exposure measurements are required. The upper percentile of these exposure distributions is of particular interest, as they may experience the greatest effects on health. Furthermore, policies that aim to reduce mean population exposure may not necessarily address the specific exposure sources that affect this upper percentile group (Edwards et al., 2005; Edwards and Jantunen, 2009). Hence, analysis of what we term the Population Exposure Frequency Distribution (PEFD) is central to understanding the variable patterns of exposure in different locations, and within different population groups, and the impact of different policy interventions.

In order to assess the benefits of policy interventions both indoors and outdoors in terms of the PEFD, a modelling approach is needed that integrates personal activity profiles with models of indoor concentrations, and specifically those within the home environment, as well as data on outdoor concentrations in different locations(e.g. Wu et al., 2005; Physick et al., 2011). In most western societies, people typically spend over 90% of their time indoors, much of it at home, and therefore understanding of the processes controlling indoor concentrations, and their relationship to outdoor concentrations, is essential if personal exposures are to be predicted with any degree of certainty.Various indoor exposuremodelling techniques are available,rangingfromsimple statistical regression and massbalanceapproaches, to more complex multi-zone and computational fluid dynamics tools (e.g. Fabian et al., 2012; Milner et al., 2011).

Over the past two decades, several probabilistic micro-environmental models of population exposure across a city have been developed, including the probabilistic SHEDS-PM model, which estimated population distributions of PM2.5 exposures in Philadelphia (Burke et al., 2001),the models derived from the European EXPOLIS study (e.g. Kruize et al., 2003; Hanninen et al., 2003), and a recent probabilistic exposure model that predicts distributions of children and elderly PAH exposures in Rome (Gariazzo et al., 2015).However, these models generally use an empirical approach to modelling indoor concentrations, and do not explicitly simulate the effects of specific indoor sources. For example, Borrego et al. (2009) predicted spatial patterns of human exposure to ozone, NO2 and PM across Portugal using a single indoor-outdoor ratio to predict indoor concentrations; such an approach cannot account for the wide variation in indoor concentrations depending on differing building characteristics, time of day, and indoor emission sources. More recently, probabilistic models of residential concentrations of NO2 and other pollutants across a city have been described (e.g. Logue et al., 2014), whichtypically treat the home as a single micro-environment using a mass balance approach; however, NO2 concentrations in kitchens, bedrooms and living rooms vary significantly, as does the time spent in each room by different population groups.

The aim of our study was to develop and apply anovel population exposure model (EXPAIR) to simulate the population exposure frequency distribution (PEFD) to NO2, and its variation with time; unlike other models, we aimed specifically to simulate the effect of indoor sources in different rooms within the home and on population groups with contrasting time-activity profiles.We also aimed to assess the effect of policy interventions both outdoors (i.e.reductions in outdoor NO2 concentrations within air quality management areas) and indoors (i.e.reduced indoor emissions from gas cooking)on modelled distributions of population exposures,using data from the UK city of Leicester.

2. Methods

2.1 Overview of exposure model (EXPAIR)

The probabilistic population exposure model EXPAIR combines the outputs from the probabilistic micro-environmental model INDAIR-2 with the time-activity-location profiles of population groups within a city. Thus the inputs to EXPAIR are:-

  • Air pollutant concentrations in the MEs in which this particular group is located, throughout the day, as generated by the microenvironmental INDAIR-2 model, divided by day of the week (weekdays/weekends) and season (summer/winter). See below, and Section 3.1 for the specific methods for generating these profiles.
  • Time-activity-location patterns over the course of a day for each population group, divided by day of the week (weekdays /weekends), and season(winter and summer), for the selected micro-environments (MEs) (seeSection 2.2).

Both location of the population groups, and outdoor air pollutant concentrations, which are used as input to INDAIR-2 simulations, are defined for different zones within a city, depending on land use and traffic density; for the application described in this paper, four different zones were differentiated (see Section 3 for details).

For the simulations reported in this paper, the simulation period was 24 hours and the time-step was 15 minutes. Model simulations follow the following stages.

1. A library of ME concentrations isgenerated by the new micro-environmental probabilistic INDAIR-2 model (see Section 2.3 for details).INDAIR-2 simulates residential MEs (assuming either no sources, or cooking and smoking activity) and non-residential MEs, simultaneously. In these runs, the pollutant concentrations in each ME are simulated at each time step from outdoor concentrations and indoor activity, using regression coefficients which are defined as a frequency distribution. Outdoor concentrations at each step are also defined as a frequency distribution for the relevant season and day of the week. Using a large number of iterations (1000, to stabilise the results), the INDAIR-2 model randomly generates values from these distributions to predict the pollutant concentrations in each ME. Based on these results, a library with the daily profiles of the air pollutant in each ME is constructed, expressed as a frequency distribution (mean, s.d.), at each 15-min interval. These daily air pollutant profiles are then used as inputs into the EXPAIR personal exposure model.

For a city specific application, the INDAIR-2/EXPAIR modelling framework allows the different outdoor concentrations, in different zones across the city, and the mobility of the population across zones during the day, to be accounted for. In the application described in this paper, we used an analysis of traffic density and fleet composition on the urban road network to distinguish four distinct zones with different outdoor NO2 levels (see Section 3.1), but other methods of defining zones could be employed. We used outdoor measurement data for one particular year, although multi-year data or modelling predictions could also be used as input to INDAIR-2/EXPAIR. Within each zone, outdoor concentrations, for each day of the week and each season, were defined for each hour as the frequency distributions of all the relevant days with measurement in that year.

2. EXPAIR is then run for the time-activity-locationprofiles (see Section 2.2) of the relevant population group, season and day of the week, to produce frequency distributions of exposure (PEFDs) across the population at each time-step. At each time-step, EXPAIR randomly selects 1000 pollutant concentrations from the INDAIR-2 concentration frequency distributions. The number of times that EXPAIR is linked to each ME depends on the time-activity-location profiles, which give the percentage of the simulated population group occupying this ME, at a particular time-step. For example, if at midnight (00:00h), for the population group of children, 95% were in the bedroom, 1% were in the kitchen and 4% in the lounge, then for a simulation of 1000 iterations, the EXPAIR model links 950 times to the concentration frequency distribution of the bedroom, 10 times to the concentration frequency distribution of the kitchen and 40 times to the concentration frequency distribution of the lounge.Thus, at each 15-min time-step, PEFDs are generated for each population group.

3. The procedureabove is modified to consider city specific time-activity-location profiles, which take account the zonal variation in outdoor pollutant concentrations, and in the location of the MEs that are used by the population group, to quantify the city-wide PEFD. The outdoor concentration profiles that are defined separately for each zoneare used to generate separate INDAIR ME libraries for each zone,while the percentage of each ME within the zone is also quantified (as described in Section 3.2 for the specific application). The selection of MEs by EXPAIR is then modified accordingly. For example, if in the above example, 50% of homes were in zone 1, 50% of homes in zone 2, and none in zones 3 or 4, EXPAIR at each time-step would link 475 times to the bedroom ME profile for zone 1, 475 times to the bedroom profile in zone 2, and would not select any values from the bedroom profile in zones 3 and 4.

2.2Time-activity-location profiles

The general time-activity profiles of three population groups (children aged 4-15, office workers, and adults aged 65+) were derived from the BBCSurvey for Daily Life (Telmar, 1998), a comprehensive database of activity for the UK population, in whichactivities and locations were recorded in the previous 24 hours, in half-hourly segments (quarter hours between morning and evening peak times). This allowed us to define six non-residential MEs (outdoor, transport, school, office, shops and large buildings, bars and restaurants) and three residential MEs (kitchen, lounge, bedroom). Since this survey was conducted for different purposes, assumptions had to be made in constructing the model location categories from location and activity data within the survey. In particular, the proportions of population in the 'home' location of the BBC survey were split into the residential microenvironments (kitchen, living room, bedroom) using combinations of appropriate BBC survey activity categories (e.g. asleep, childcare, housework, watching TV, preparing food). The derived time-activity profiles for the three population groups for winter weekdays are illustrated on Figure 1.

The city specific time-activity-location profiles were created from the general activity profiles based on land use and traffic density. The distribution of homes, and the five indoor non-home MEs, between the different zones, as distinguished above for the outdoor NO2 concentrations, can be defined in a number of ways, depending on the availability of GIS data, location information, and transport use statistics. In this specific application, we used a mix of methods, as described in Section 3.2.