Introduction

An individual’s exposure to urban air pollutants is primarily dependent on the environments in which they spend their time, along with the quality of the ambient air in which these environments are positioned within. Hazardous microenvironments and “hot spot” zones throughout cities cause exposure between individuals to fluctuate substantially, even if they happen to be neighbors or live within the same household. This poses an interesting challenge for exposure science and epidemiology. A single fixed site monitor is an insufficient indicator of exposure for individuals living throughout a wide urban area. Ideally, air pollutant measurements, location, and activity data are recorded as the person moves through time and space, but this is impractical for studies assessing health data across large populations. Consequently, crude exposure estimates are generated and linked to health outcomes, often at the census area level.

Current preferred techniques include land use regression (LUR) and hybrid models, which incorporate multiple modeling aspects from a wide range of inputs. Although ideal for large numbers of individuals, hybrid approaches almost always rely on one or more datasets to be artificially generated, which increases uncertainty, possibly enhancing exposure misclassification, e.g., emissions dispersion, concentration interpolation, land use activity, or extrapolation of population inputs. The most reliable studies go one step further by validating and/or “training” (refining) the hybrid model with real-world time-activity and microenvironment exposure data (personal measurements), but these are generally limited to a small number of individuals.

A recent example came from Münster, Germany, where good agreement between modeled and measured PM10 and PM2.5 exposure was achieved for a study of ten participants using a global positioning system (GPS), activity diaries, and ambient data, despite excluding monitored indoor microenvironment concentrations (Gerharz et al. 2013). Another study in Lieria, Portugal, estimated personal PM2.5 exposure using GPS data collected by five individuals, in combination with traffic emissions and dispersion models. Although they did not attempt to validate the modeled exposure levels, they compared well with those from similar studies in other European cities. The validation of model output is desirable but not always required, depending on the objectives of the study. Smaller studies which model exposure for just a few individuals have the advantage of greater accuracy yet there are limitations regarding statistical representativeness and applicability to a wider population. Accordingly, estimated exposures may not be useful in analyses of health census data, unless the results can be reliably extrapolated to wider areas.

Since the 1990s, the US Environmental Protection Agency (US-EPA) has released a range of large-area population exposure models. The most recent of these is the Air Pollution Exposure Model (APEX), which is designed to run exposure and/or inhalation (dose) simulations for large populations (groups of individuals) as they move across space and time, throughout a variety of microenvironments. Due to the ability to account for the movement of individuals, it is a significant improvement on traditional LUR methods and is consistently cited by the scientific literature as being one of the key human exposure models available (Fujita et al. 2014; Georgopoulos et al. 2009; Isaacs et al. 2008; Özkaynak et al. 2008, 2013; Xue et al. 2010; Zou et al. 2009).

APEX was recently applied and evaluated in a large-scale, multi-model study for the whole of Atlanta GA, USA (Dionisio et al. 2013). Modeled daily exposures to NO x and CO were found to be in good agreement (r > 0.82) with those from hybrid models tested in the study, demonstrating the viability of using APEX as a standalone option. Dionisio et al. (2013) go on to state that exposure models will likely hold an advantage in fine-scale spatiotemporal applications and note the possible suitability for epidemiological investigations. However, a subsequent health effects-exposure investigation (cardiorespiratory emergency visits) did not produce stronger estimates of effect than those from ambient monitoring data, with the exception of the association between NO x /CO and asthma/wheeze (Sarnat et al. 2013). Despite these results, it is clear that spatiotemporally resolved personal exposure data have a future role in this area and associations will likely improve with model refinements and increased activity diary accuracy.

In this study, we use APEX to simulate the exposure of 51 near-highway residents to traffic-generated (NO x , CO) and background source (PM10) pollutants. Unlike previous population exposure studies covering wide areas, we focus on individuals living within close proximity to one another across a community no larger than 1.5 km2. The community is largely comprised of low-income, ethnic minority groups, and household mould, overcrowding and a lack of adequate heating are common issues (Bullen et al. 2008; Butler et al. 2003; Cheer et al. 2002). Pre-existing health conditions are further compounded by exposure to emissions and dust resuspension from a large highway running through the community. Asthma, diabetes, and heart disease are of primary concern, and these afflictions are disproportionately high within the local population compared with the rest of the city (Cheer et al. 2002).

Time-activity profiles were derived from our own local surveys and then divided into eight different occupational groupings, based on the proportion of time spent in common microenvironments. We also use continuously monitored air quality data from Federal Equivalent Method (FEM) instrumentation at several sites across the study area, as opposed to concentrations approximated from emissions or air dispersion models. The effect of proximity and position of the home location relative to the highway is explored within the context of time spent away from the home, i.e., occupation, along with the proportion of time spent outside, are the main drivers of varying exposures. In addition, we shift the population around the study area for subsequent model runs in order to quantify the impact of proximity of the home. Much of the near-highway literature concludes that concentrations of primary traffic emissions (NO x , CO, ultrafine particles) are typically elevated by approximately 50 % at the roadside compared to 100–150 m downwind (Karner et al. 2010; Pattinson et al. 2014; Patton et al. 2014). For long-term health reasons, some studies are now recommending a minimum separation of all residential buildings from highways, e.g., 100 m (Barros et al. 2013), but little is known about who would actually benefit most from a separation of such a distance.

Our aims are threefold. Firstly, we aim to simulate and compare exposures for worker and nonworker profiles assigned to a range of occupations. Secondly, we explore the effect of placing the study population directly within the roadside corridor (highest exposure zone) compared to further back from the road. We refer to this as the effect of proximity and position, relative to the highway and the direction of local winds, respectively. Thirdly, we assess which occupational groups would obtain the least and most benefit (estimated reduction in exposure) if their home was to be shifted a minimum of 100 m in either direction, out of the highway corridor.

Our study is novel in that we take a population exposure model and apply it to a fine-scale study on the exposure impact of residential proximity to a major emissions source. We also believe that it is the first personal exposure modeling study conducted for New Zealand.

Methods

Study area and air quality monitoring

The study was primarily conducted in Otahuhu, South Auckland, a suburb bisected by a six-lane highway for which Annual Average Daily Traffic (AADT) exceeds 120,000 vehicles (NZTA 2013). The local monitoring campaign featured three air quality and meteorological stations at 5 m downwind, 134 m downwind, and 212 m upwind of the highway, relative to the predominant southwest wind direction (Fig. 1). This layout represents a typical near-highway monitoring configuration, with the three stations providing one immediate roadside site (5 m downwind), one downwind background site (within the potential spatial extent of highway emissions) and one upwind background site (likely outside the potential extent of highway emissions). Each monitoring station was equipped with instrumentation to continuously measure NO x (chemiluminescence analyzer; model 200, Teledyne API), CO (gas filter correlation analyzer; model 300E, Teledyne API), PM10 (BAM; Model FH62C14, Thermo Scientific) and wind data (anemometer; Models A101M & W200, Vector Instruments) at a time resolution of 10-min mean values. In the interest of capturing a wide range of pollutant concentration levels, which included coarse particulate from wood smoke, air quality sampling took place during a mid-winter month; specifically, midnight of July 1 to midnight of July 31, 2010. The dataset used in this study is a subset of a long-term highway corridor monitoring campaign, detailed in a separate article (Longley et al. 2014).

Fig. 1
figure 1

Study area showing wind direction and speed (m/s), location of monitoring stations, actual locations of study participants’ homes, and simulated participant home locations, relative to the highway. Note that actual resident locations are inexact to preserve anonymity (randomly shifted 1–2 properties either side of home address)

Full size image

In addition to the primary study area, data from five other monitoring sites were used to represent residents’ work locations (Fig. 2). This gave a total of six work locations (including Otahuhu itself) represented by a local ambient air monitor (Fig. 2). Data for the two monitoring sites north of Otahuhu, Penrose and Grafton (Fig. 2), were sourced from the Auckland Council which operates identical models of instrumentation to those deployed by our study (Auckland Council 2005). As the local council does not have other monitoring sites in the vicinity of South Auckland, the other three destinations of East Tamaki, Papatoetoe, and Mangere used replicate data from the upwind background site shown in Fig. 1. Note that all three sites are similarly positioned approximately 215 m upwind from a major road (Fig. 2). Essentially, we can either make the assumption that our upwind background site represents levels that would be found at a similar site within the neighboring suburbs or that the local worker does not leave their home suburb. The purpose was to be able to replicate actual time-activity-location patterns as closely as possible, within the exposure model.

Fig. 2
figure 2

Map showing wider study area and positions of other monitoring stations. The primary study area of Otahuhu is near the center of the map

Full size image

Participant survey methods

To gain a realistic understanding of how local residents actually spent their time, time-activity data was required for the model. Due to a lack of existing time-activity data for Auckland at the temporal resolution and level of detail required, we opted to conduct our own survey of local residents by means of door-to-door recruitment. Prior to conducting the surveys, the survey was reviewed and approved by the University of Canterbury Human Ethics Committee (HEC 2011/94). The aim was to recruit 25 participants from within the immediate roadside corridor (<150 m down/upwind) and 25 outside of the corridor, yet still within potential influence of the highway (>150 and <500 m down/upwind). Some residents asked the interviewer to return at a different time. In keeping these appointments, the number of participants exceeded the target by one. Each participant was asked to provide a timeline of their movements for a typical week. Targeted door-to-door recruitment for the purposes of time-activity surveying has been used in previous exposure modeling research (Brugge et al. 2013; Kaufman et al. 2012). With the exception of short trips within the local area (15 min), each activity lasted a minimum of 30 min. For most participants in full-time employment, their Monday to Friday schedule was generally identical but punctuated by regular sports practice, club meetings, shopping, and social outings on particular evenings. The majority of part-time workers and the unemployed or retired also had routine schedules that generally revolved around child care (school times) and/or regular recreational activities. Common weekend activities included attending church, visiting friends/family, and local recreation such as sports and fishing. None of the participants routinely left the wider study area, and we were not interested in factoring in atypical movements such as long-distance travel.

As this study only focused on exposure and not inhaled dose, it was not necessary to gather sensitive demographic data such as age, height, and weight. However, participants were asked to provide their work location, type of employment and type of work environment (building and ventilation type), as well as the approximate location (the suburb) of any regular activities outside of work. These data were used in combination with the time-activity diaries to assign the range of microenvironments and air monitoring locations that participants passed through during a typical week.

Exposure model inputs and configuration

APEX allows for a highly complex variety of model inputs which can be used to estimate exposure and pollutant dose for individuals living and working in numerous types of microenvironments. Outcomes are dependent on a wide range of user-input environmental (ventilation systems, presence of gas utilities, air conditioning, window use, etc.) and physiological parameters (age, gender, weight, height, etc.). We omitted much of this information as it was not required for this application, e.g., physiological parameters are used in the calculation of dose, but not for exposure.

At the most basic level, APEX assigns an average exposure to each environment/microenvironment based on time spent in the environment and estimated concentrations within the environment, which are derived from an calculated concentration of indoor and outdoor sources, based on penetration factors and air exchange rates (AERs). Local air quality and meteorological data from multiple stations is needed as the primary model input, along with time-activity information.

The following equations show how APEX calculates concentrations in different microenvironments:

$$ {E}_{\mathrm{tot}}={\displaystyle \sum_i{f}_i*{C}_i} $$

where

E tot :

total exposure, expressed as an average concentration for the exposure period

f i :

time fraction spent in the microenvironment

C i :

concentration in the microenvironment

Exposure for individual microenvironments are calculated as follows:

$$ {C}_i = {C}_{\mathrm{out}}*{p}_i+{S}_n $$

where

C i :

concentration within the microenvironment

C out :

outdoor concentration at microenvironment location

p :

pollutant penetration factor

S :

attribution of sources in the indoor microenvironment

APEX was run without the inclusion of pollutant concentrations originating from indoor sources, or S. As residents have control over sources originating from inside their homes but little to no influence on ambient air quality, including indoor sources was not an objective of this study.

In addition to our own local pollutant and meteorological data, we added air exchange rates (AERs) measured inside a 1960s weatherboard home with the windows and doors closed, situated next to the downwind background station (Fig. 1). AERs were measured by continuously logging the decay of carbon dioxide (CO2) concentrations (model GMP343, Vaisala) after releasing a set of 12 g CO2 cartridges. This technique has been developed and validated both in the lab and within field studies (You et al. 2012). Five AER measurements were made over a 2-week period during the pollutant monitoring campaign. The range of AERs (1.4–2.8 ac h−1) was then applied to the residential microenvironments for all individuals, across all days within our study. Although it would have been ideal to measure within both older and newer homes, the type of dwelling used is most representative of the housing stock within Otahuhu and throughout other low-income areas of South Auckland. AERs for larger building types, as well as penetration factors for vehicle types, were left to the default parameters based on data from US cities (see Appendix 1 for full table of parameters).

Exposure model application

APEX is designed to run large simulations for thousands of profiles (individuals) across wide urban areas. It is a stochastic model, where individual profiles (demographic and time-activity data) are randomly assigned to area units (sectors) within the defined study zone. Similarly, a value for each parameter relating to each microenvironment is randomly assigned, based on a range of pre-determined values. For example, no two buildings will have exactly the same penetration factors and air exchange rates, but similarly-sized buildings will likely fall into a range of known parameters (previously measured) for that type of building and ventilation system. Most applications of APEX use pre-existing human activity databases in combination with known microenvironment parameters. When APEX is run in this way, an unlimited number of profiles can be generated, with the results reflecting random scenarios (different activity patterns in varying microenvironments), for the same individual, over the study period.

We removed the stochastic profile element by assigning each individual diary day, to each participant’s exact geographic location at home and work (sectors). The same initial random number seed was used for all model runs, eliminating possible confounding by stochastic elements when comparing different runs. The only random element remaining was the assignment of parameters to each microenvironment. As the order of simulated profile output is also randomized, this was controlled for by generating 255 profiles (4 for each person) and using the results for the first instance of each profile, across all model runs. Figure 3 illustrates the way in which APEX calculates concentrations in order of sequence.

Fig. 3
figure 3

Flow diagram showing sequence of APEX model process

Full size image

Appendix 2 provides an example of the detail of diary input into APEX for a worker who wakes at 0530, spends 30 min getting ready and having breakfast and takes a 15-min commute before commencing work at 0615. Each time a microenvironment changes, a different activity and location code apply. These included both indoor and outdoor environments in private, residential settings, at workplaces, and in publicly accessible areas such as parks and shopping malls. A complete list of the microenvironments used (17 of 28 available) and the concentration estimation method employed for each is provided in Appendix 3. For 51 persons over 30 days, assigning activity and location codes required transforming written diary data into 10,502 lines of code, resulting in an exposure simulation of 1530 individual person days. Summary statistics for percentage of time spent in different activities and environments are provided in Appendices 4 and 5, respectively. The simulation was run once using the residents’ home locations, and four times with the population artificially placed 50 and 150 m downwind, as well as 50 and 150 m upwind, relative to the highway (refer to Fig. 1). Although APEX includes a module to simulate commuting, we did not include exposures while commuting in our study as the focus was on ambient exposures and the impact of near-highway living.

Analysis of model output

The resultant mean exposure value for each individual (30-day simulation) was pooled into one of eight occupational groupings, based on time spent in common microenvironments, such as ventilated offices and outdoor work environments. All occupational groupings are self-explanatory with the exception of the unemployed or retired group. These profiles were divided into two groups, one being active and the other inactive. An unemployed or retired participant who spent more than 70 % of their time inside their home (not going outside or leaving the property) was defined as inactive (see Appendix 5). This gave a range of 60–360 simulated exposure days for each group (2–12 individuals).

Additional analyses on the daily exposure data were performed by applying a generalized linear model (GLM) to the eight groupings. The primary model tested proximity and position of home location, along with occupational group, as the main effects (independent variables). Post hoc Bonferroni tests were used to assess significance between means for all possible pairings by occupation and position relative to the highway, i.e., pairs both within and between exposure groupings. A second GLM tested the effect of work location on exposure. A single GLM could not be used alone due to a lack of variance across categorical variables, e.g., all unemployed/retired profiles have the same work location of “Home.” For clarification, all variables used in the GLM analyses are provided in Appendix 6. To check for any bias within the models, relationships between all variables were tested within a correlation matrix. All analyses were tested for significance at alpha level 0.05 but are reported as <0.05, <0.01, or <0.001, as per the results. The analyses were run within a statistical software package (Statistica 10, StatSoft Inc.).

Results

Fixed station measurements

Table 1 provides the summary statistics for all local and work destination monitoring sites for the duration of the modeling campaign (refer to Fig. 2 for geographical locations). Within the local study area, the monitor 5 m downwind (representing the roadside corridor) recorded concentrations ~50 % (NO x ) and ~43 % (CO) greater than each of the sites further back from the road, which were almost equal. Despite having considerably less traffic capacity (four lanes instead of six), measurements at the street canyon site were 25 % (NO x ) and 30 % (CO) higher than at the highway roadside in Otahuhu.

Table 1 Fixed station mean results from hourly measurements for July 2010
Full size table

Figure 4 illustrates the strength of the street canyon effect, with the Grafton site showing a clear bi-modal peak representing the rush hour traffic flows. The late rush hour peak at the highway area is suppressed by strong afternoon winds. Although there is a significant emissions source at Otahuhu, it is heavily modulated by local meteorology. This is important in the context of utilizing residents’ time-activity-location dairies, because there are major diurnal fluctuations in the spatial extent of traffic markers (wider during mornings and evenings). For a full discussion on spatial extent in this area, see Pattinson et al. (2014). Variation in PM10 between sites was minimal.

Fig. 4
figure 4

Diurnal NO x concentrations closest to the roadway, July 2010

Full size image

APEX model output: exposure estimates

Proximity to highway

For resident proximity to highway (both sides of highway), there was an inverse relationship for NO x (r = −0.28) and CO (r = −0.32), but no trend for PM10 (r = 0.00). When residents’ actual home locations were shifted to the immediate roadside corridor at the downwind edge (50 m), exposure for the group increased by 31 % for NO x (F 39 = 10.3, p < 0.001) and 36 % for CO (F 39 = 10.7, p < 0.001). However, when downwind separation was extended to 150 m, exposure decreased by 56 and 71 % from the 50-m point, respectively (Table 2). Mean differences at the upwind side were only statistically significant at the 150-m point. Exposures ranged from 31 to 48 % less than 50 m within the downwind corridor or from the residents’ actual locations (Table 2). Note that 13 of the 51 simulated profiles reside within 50 m downwind of the highway (see Fig. 1).

Table 2 Total group differences in ambient month-long exposure for July 2010 (1500 individual profile days simulated)
Full size table

Proximity to highway and occupational group

The main effects on exposure were highly significant (p < 0.001) for all three pollutants (Tables 3 and 4). Proximity and position were important, as was occupational group, irrespective of where participants lived. A basic sensitivity analysis, where the three groups of the smallest sample sizes were removed, showed that the main effects remained highly statistically significant at p < 0.001. Additionally, there was no correlation between home position and occupational group (r = −0.06).

Table 3 GLM multivariate tests: proximity to highway and occupational group
Full size table
Table 4 Whole model R: proximity to highway and occupational group
Full size table

Figures 5, 6, and 7 illustrate the varying levels of exposure grouped by occupation and proximity/position relative to the highway. Figure 8 provides an illustration of day-to-day NO x variation throughout the study period, for each occupational group. For all three pollutants, outdoor laborers were the most exposed group and the unemployed or retired, inactive group, the least exposed. While variation within the results for PM10 was minimal, post hoc testing revealed multiple statistically significant differences within and between groups for NO x and CO (Appendices 7 and 8). Those who spent more time doing outdoor activities at home, or more time near traffic at work, faced increased exposure at the 50-m downwind position. NO x exposure grew 47 % (F 229 = 10.3, p = 0.005) and 93 % (F 229 = 10.3, p = 0.021) for the unemployed/retired-active and security guard groups, respectively (Fig. 5 and Appendix 7). For the same occupations, CO exposure increased 53 % (F 229 = 10.3, p < 0.001) and 157 % (F 229 = 10.3, p = 0.004). All occupations benefited from a further 100 m separation from the highway, with 53–75 % reductions for both pollutants (p < 0.05). Again, the most substantial reductions were seen for the unemployed/retired-active and security guard groups. Across all groups, the range between the least exposed (unemployed/retired-inactive) and most exposed group (outdoor laborer) varied by a factor of 6 and 8 (NO x , CO) from 50 m downwind to 100 m downwind. This shows that persons living just 100 m from one another can potentially have an eightfold variation in ambient exposure. Complete summary statistics for the occupational groups by proximity and position can be viewed in Appendices 7, 8, and 9.

Fig. 5
figure 5

Simulated total mean ambient NO x exposure by occupational group for the month of July, 2010. Error bars denote 95 % confidence limits

Full size image
Fig. 6
figure 6

Simulated total mean ambient CO exposure by occupational group for the month of July, 2010. Error bars denote 95 % confidence limits

Full size image
Fig. 7
figure 7

Simulated total mean ambient PM10 exposure by occupational group for the month of July, 2010. Error bars denote 95 % confidence limits

Full size image
Fig. 8
figure 8

Variation in simulated daily mean NOx exposure by occupational group for the month of July, 2010

Full size image

Proximity to highway and work location

An additional GLM was run to test the effect of work location (Tables 5 and 6). Mean exposure values by work location are illustrated in Fig. 9. Although there was a moderate strength statistically significant correlation between work location and occupation (r = 0.41, p = 0.03), this was only due to the unemployed/retired group all having the same work location of “Home.” With this occupational group removed, the correlation ceased to exist (r = 0.006) yet the main effect of “Work Location” remained significant at p < 0.001 within the GLM. Thus, we can be confident that work location has a significant influence on the simulated profiles, independent of occupation.

Table 5 GLM multivariate tests: proximity to highway and work location
Full size table
Table 6 Whole model R: proximity to highway and work location
Full size table
Fig. 9
figure 9

Simulated mean ambient pollutant exposure by work location for the month of July, 2010

Full size image

Discussion

Monitored ambient levels versus modeled exposure

The mean values recorded at the nearest associated monitoring station were a factor of 2.3–2.8 (NO x , PM10) times greater than the overall mean for the grouped (by proximity) exposure profiles. For CO, this divergence was even greater at 2.5–4.8 (Tables 1 and 2). These basic results alone clearly highlight the protective effect of the residential building envelope, and the importance of time spent in other environments, away from home. Even with relatively high air exchange rates, naturally ventilated environments can help to downscale indoor levels to half those of outdoor, when no indoor emissions sources are present (Jones et al. 2000; Ní Riain et al. 2003). Although indoor sources are important for personal exposure assessment, indoor residential sources of NO x and CO are typically limited to unflued gas appliances, fireplaces, and tobacco smoke. Conversely, ultrafine particles (UFPs) are emitted in very high numbers during numerous household activities, especially cooking (Buonanno et al. 2014). Wood burners, other indoor combustion activities and resuspension from household movement can contribute to PM10 levels, but in this case, we are only interested in the impact of being in close proximity to a highway. The rationale for this is that the generation of indoor sources at home is something that can be controlled by individuals whereas outdoor air quality cannot. The pollutants chosen for our study are appropriately suited for representing fresh exhaust emissions (NO x , CO), traffic-generated coarse particulate, and the urban background plume (PM10).

Overall effect of proximity

The overall effect of proximity to the highway on mean NO x and CO exposure was substantial. If the study population were housed within 50 m downwind of the highway, exposure would be 32–37 % greater than when dispersed throughout the community (actual locations, Fig 1). If the separation was increased a further 100 m outside of the immediate downwind corridor, exposure would decrease 56–70 % (Table 2). Exposure at the upwind side of the corridor was also considerably less (16–18 %). The roadside to downwind background (134 m) decrease from the monitored data was lower at 51 and 44 %, for NO x and CO, respectively (Table 1). One factor explaining this difference may be time spent outside (by active individuals) elevating the roadside group mean, increasing the gap in simulated exposures.

The majority of roadside studies report an approximate 50 % drop in NO x and CO within 150 m, but no decay trend for PM10 (Karner et al. 2010). Our findings are in agreement and also show that when time-activity-location patterns are taken into account, ambient exposure reductions can potentially increase. This has important implications not only for planning and development, but also for social housing policy in regard to the placement of sensitive individuals and groups. The long-term impact of living within the highway corridor can be considerable for children, the elderly, and those suffering from significant illness and/or respiratory afflictions (Favarato et al. 2014; Perez et al. 2013). This is a concern within the local study context as its population suffers from disproportionately high rates of respiratory disease and diabetes (Cheer et al. 2002). Immediately south of Otahuhu, there is a primary school within 10 m of the highway which also features an early childhood center on school grounds. For some residents living in South Auckland roadside communities, time spent away from the highway corridor could be quite limited.

Effect of occupation by proximity

To the best of our knowledge, no previous exposure modeling simulation has explored the influence of occupation, whereas personal monitoring by different occupations is more commonly investigated. Most of these tend to focus on high-exposure occupations such as those who work within heavily trafficked zones (policemen, taxi drivers, toll gate employees) and potentially toxic workplace settings (commercial cleaning operations, beauty salons, autobody repair workshops) where volatile organic compound (VOC) vapors and aerosols can be present in high concentrations (Bello et al. 2009; Kisku et al. 2013; Tsigonia et al. 2010). Very few personal monitoring studies exist for NO x and those for CO usually include significant indoor sources (di Marco et al. 2005; Raw et al. 2004). For better comparability, we refer to research utilizing markers of fresh traffic emissions which do not have strong within-residence sources.

An occupational study of benzene (a highly carcinogenic VOC present in petrol fumes) exposure for 50 workers in Athens, Greece, found the strongest predictors were proximity of home location to heavy traffic, time spent outdoors, and time spent in transportation (Chatzis et al. 2005). Time spent outdoors explained the link between occupation and high personal exposure levels. These findings are reflected by our results, with the outdoor laborers and security guards consistently the most exposed, while professionals, office workers, teachers, and students were the least exposed (Figs. 6, 7, and 8). Similarly, exposure for those who spent more time outdoors at the home location (unemployed/retired, active) was 43–45 % (NO x , CO) and 46–50 % (PM10) greater than for those who spent most of their time indoors. For these groups, differences for NO x and CO were only significant (p < 0.05) within the roadside corridor (both sides) but shifting location did not render PM10 nonsignificant anywhere. Similar levels of significance, again for the roadside corridor only (NO x , CO), were found for other paired groups such as outdoor laborer versus indoor laborer. Those working in large warehouses and factories were 28–37 % (NO x , CO) and 38–40 % (PM10) less exposed than carpenters working outside. Further back from the highway, the only occupation for which exposure was statistically significantly (p < 0.05) higher for NO x and CO was outdoor laborers, and only when compared to those who spent most of their time indoors at home (unemployed/retired-inactive). With the exception of PM10, these findings show that when living outside of the highway corridor, the effect of occupation is nonsignificant for most but remains somewhat important for those who may be working outside near busy roads.

These results also show that proximity of residence has a far stronger influence on personal exposure to ambient pollutants than occupation. A VOC study of 100 residents in London, UK, found that 50–75 % of variability within personal exposure for most compounds was explained by the home environment (Delgado-Saborit et al. 2011). Extensive source apportionment work in Helsinki, Finland, concluded that traffic emissions were second only to domestic cleaners, as the strongest origin of indoor VOC concentrations in nonsmoking households (Edwards et al. 2001). Another urban exposure project in Camden, USA, found that ambient concentrations of polycyclic aromatic hydrocarbons (PAH), from traffic emissions and industrial combustion, explained 44–96 % of variability in personal exposures (Zhu et al. 2011). We can therefore be fairly confident that, in the absence of significant indoor and/or occupational sources, ambient data can be used in personal exposure estimation to obtain meaningful results. Much of the remaining variability for these studies is generally explained by exposures while commuting.

A further outcome from our analysis is that we have gained an indication of how differing occupation profiles would potentially benefit from the >100 m roadway separation recommended by recent literature (Barros et al. 2013; Hystad et al. 2013). Likely owing to the weaker presence of northeast winds blowing toward the leeward side (27 % of all observations), increasing the separation of the group from 50–100 m was only significant (p < 0.05) at the windward side (downwind for 58 % of hourly observations, Fig. 1). The mean decrease downwind was 56 % (NO x ) and 71 % (CO) for the group and some occupations benefited more than others, but not by a great margin (5–7 %).

Perhaps a more important finding is that the unemployed/retired-active and security guard groups, who spent a large proportion of time outdoors and near traffic sources, would face a substantially increased exposure burden if living within the downwind corridor than if living elsewhere. Those working in ventilated offices, who spend less time outdoors for work and/or recreation, are afforded better long-term protection from ambient air. This is indicated by the results for those working in the CBD area (Grafton), who spent 8 h of each weekday inside mechanically ventilated office buildings. Despite being in the most polluted area (an inner-city street canyon), NO x exposure was lower than for those who worked at four (out of six) alternative locations, in different environments (Fig. 9). This finding is supported by previous microenvironmental comparisons for office workers in Los Angeles, USA (Fujita et al. 2014) and Hertfordshire, UK (Kornartit et al. 2010). Comparatively, those working near traffic or outside are faced with higher exposure both during the day and then again when at home, especially if active with gardening and recreation in the evenings and during weekends. Gardening or tinkering in the yard was a regular activity for two thirds of our unemployed/retired-active participants. For our security guards, both were positioned outdoors near busy roads. All occupations which involved spending a lot of time outside and/or near traffic resulted in higher exposure profiles than the unemployed/retired-inactive group, whose members collectively spent 85 % of their time indoors at home (Fig. 8).

Recent exposure modeling studies

Moving away from population-based, citywide (mesoscale) simulations which utilize generated emissions data and/or proxy variable inputs, is likely to substantially improve the accuracy of personal exposure estimates. These individual results can thereafter be collated to build results for a wider population, which can form a stronger basis for epidemiological studies. One example originates from Münster, Germany, where researchers kriged ambient and local particulate concentrations over a 250-m2 citywide grid, included GPS time-activity data, indoor emission sources, and indoor/outdoor ratios for transport microenvironments, achieving strong correlations between modeled and measured exposures (Gerharz et al. 2013).

An NO2 study using LUR methods in Antwerp, Belgium, found a mean exposure range of 11–36 μg/m3 (Dons et al. 2014b). Assuming a kerbside NO2/NO x ratio of 31 % (Wang et al. 2011), the estimated NO2 portion for our roadside corridor mean individual exposures would closely match this range at 9–31 μg/m3 NO2 or 29–104 μg/m3 NO x (Appendix 7). A mixed-method study trialing a range of kriging and regression modeling techniques in the UK calculated population mean results of 54–70 μg/m3 for Blackpool and Manchester, respectively (Hannam et al. 2013). The best of these performed fairly well against results from personal passive samplers (r = 0.60–0.62). Further research in Hillsborough County, FL, USA, utilizing data generated by the CALPUFF dispersion model, eventuated in an urban residential group mean exposure of 22 μg/m3 and individual exposures up to 43 μg/m3 (Gurram et al. 2014). The collective results from these NO x studies are reasonably comparable to our group means of 25–60 μg/m3 (Table 2, Appendix 7).

To our knowledge, only one publication has specifically applied APEX to a study population, in Atlanta, GA, USA (Dionisio et al. 2013). Although individual profiles were simulated, only median results at the zip code level were provided. While not directly comparable, median NO x concentrations were 0.05 ppm or ~94 μg/m3 (based on the molecular weight of NO2) and CO 0.80 ppm, considerably greater than our mean exposures, which is explained by higher ambient concentrations in Atlanta. The authors noted good agreement between median zip code APEX results and associated central site monitors even though there was substantial spread between the lower and upper 95th percentiles.

Study limitations and recommendations

The limitations within the current study are not dissimilar to those faced by previous work, with all presenting clear strengths and weaknesses. The main criticism of our work is that we used a limited sample size (n = 51), which may be statistically representative of the wider population, especially when comparing occupational groups with as few as just two members. However, smaller sample sizes are typical of this type of modeling due to the resource-intensive nature of collecting detailed time-activity data (Chatzis et al. 2005; Dias and Tchepel 2014; Gerharz et al. 2013).

Further limitations, which very few studies adequately address, include the inability to account for exposures while commuting and indoor sources at the home, as well as the lack of model validation.

The challenges associated with including commute-time exposure and indoor resident-specific sources are widely acknowledged and improvements are ongoing (Dias and Tchepel 2014; Vette et al. 2013). Exposure while commuting is currently best handled by in-traffic exposure models (Dons et al. 2014a) or LUR models that include traffic volumes/road source intensity. The inclusion of indoor sources and commuting exposure continues to be completely omitted from some projects due to associated complexities (Dons et al. 2014b; Stroh et al. 2012).

The validation of our results would only have been possible if we had the ability to sample in microenvironments with no indoor sources. This is unrealistic in the context of personal sampling where dust resuspension from human activity and emissions from cooking activity and home heating cannot be completely eliminated. Therefore, it is appropriate to omit validation in ambient modeling studies with an absence of controlled microenvironments as the results are useful in the epidemiological analysis of health effects where ambient sources are the primary interest (Dionisio et al. 2013).

Despite all of the limitations mentioned herein, our results do not deviate wildly from the ranges found in previous modeling studies (Gurram et al. 2014; Hannam et al. 2013; Zidek et al. 2005). Finally, a larger population sample would have likely strengthened statistical associations between occupational groups at the interaction effect level and improved estimates of potential exposure benefits when moving further away from the highway.

Our study has provided an interesting insight into the exposures of near-roadway residents in South Auckland and contains some meaningful results, regardless of its limitations. It is a good “first look” at personal exposure simulations for a New Zealand city and provides the foundation for future local studies, which could be expanded to focus on particular population subsets. Additionally, this unique application of a personal exposure model may help inform other highway proximity population exposure assessments in the context of time-activity patterns altering exposure. These findings are especially useful for groups and individuals with pre-existing health conditions and for those who work in occupations where exposure to traffic pollution is already high. Developing an understanding of who is affected most by near-highway living, may help steer local land use policy in a direction that aims to protect the most vulnerable citizens. Future analyses could include sociodemographics such as age and ethnicity, combined with health status data, i.e., susceptible individuals. Such an application has the potential to inform epidemiological studies, as demonstrated by previous work, e.g., Sarnat et al. (2013).