The impacts of personal mobility and diurnal ... - ACS Publications

Subscriber access provided by UNIV OF SCIENCES PHILADELPHIA

Article

The impacts of personal mobility and diurnal concentration variability on exposure misclassification to ambient pollutants Rakefet Shafran-Nathan, -- Yuval, and David M Broday Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.7b05656 • Publication Date (Web): 02 Mar 2018 Downloaded from http://pubs.acs.org on March 4, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 27

Environmental Science & Technology

1

The impacts of personal mobility and diurnal concentration

2

variability on exposure misclassification to ambient pollutants

3 4

Rakefet Shafran-Nathan, Yuval, David M. Broday*,

5 6

Faculty of Civil and Environmental Engineering, Technion, Haifa, Israel

7 8 9

All authors declare that they do not have any competing financial interests in the work

10

and its results

11 12

Key words: Air pollution; commute; exposure misclassification; integrated daily

13

exposure; time-activity patterns

14 15

Short running title: Spatiotemporal characteristics of exposure misclassification

16 17 18 19

*Corresponding Author:

20

David Broday

21

Faculty of Civil & Environmental Engineering, Technion – I.I.T., Haifa, 32000, Israel

22

Email: [email protected]

23

Tel: +972-4-829-3468

24

1 ACS Paragon Plus Environment


Page 2 of 27

Abstract

25

Appreciating the uncertainty margins of exposure assessment to air pollution requires

26

good understanding of its variability throughout the daily activities. This study

27

describes a modeling framework for estimating exposure to air pollutants for a

28

representative sample of working Israeli adults (N~168,000) for which both the

29

residence and workplace addresses were available. Individual daily trajectories were

30

simulated by accounting for five generic daily activities: at home, at work, while in

31

commute from home to work and back, and during out-of-home leisure activities. The

32

integrated daily exposure to nitrogen dioxide (NO2) was estimated for each individual

33

by tracking the daily trajectory through an NO2 concentration map, obtained using a

34

dynamic and highly resolved dispersion-like model (temporal resolution: half-hourly,

35

spatial resolution: 500 m). Accounting for the subjects’ daily mobility was found to

36

affect their exposure more significantly than accounting solely for the diurnal

37

concentration variability, yet a synergistic effect was noted when accounting for both

38

factors simultaneously. Exposure misclassification vary along the day, with the work

39

microenvironment found to contribute the most to it. In particular, regardless of the

40

high concentrations encountered during the commute, their contribution to the

41

integrated daily exposure is small due to the relatively short time spent in this activity

42

by most people.

43 44

Introduction

45

One of the challenges of a reliable exposure assessment is to integrate the pollutant

46

concentrations over the time spent in each of the various microenvironments which

47

the subjects probe along their daily trajectory 1. The most accurate way to do this is by

48

direct measurement of the personal exposure. However, in practice this approach is

49


Page 3 of 27


limited to small study groups and for relatively short periods

2, 3

, which may not be

50

representative of the whole population. Clearly this approach is not applicable to

51

retrospective studies. Moreover, since tracking the individuals’ daily activities can

52

usually be carried out only for relatively short periods, the data may not reflect

53

seasonal variation in both space and in time 2. These shortcomings might prohibit

54

achieving statistically significant results in environmental epidemiology studies

4-6

,

55

since significant results require large number of cases/individuals. Thus, many studies

56

resort to estimating the subjects’ exposure by models. In most cases, static

57

concentration maps are used

4-6

. Namely, the short term (e.g. hourly) diurnal

concentration variability is smoothed out by long-term (e.g. daily) averaging.

58 59

Reliable personal exposure estimates require detailed information on the time-

60

location-duration activity patterns of the subjects throughout the day, which if

61

combined with highly resolved pollutant concentrations in space and time can be used

62

to derive accurate exposure estimates at the individual level. In particular, the former

63

is normally related to the individuals’ commuting habits, e.g. travel mode, time of

64

day, route, and the sequence of- and time spent in each microenvironment 7. A

65

possible approach for deriving exposure estimates is to use the mean pollutant

66

concentration at each microenvironment and the list of microenvironments visited

67

throughout the day. Using this information, it is possible to approximate the integrated

68

daily exposure 5, 7, 8. Yet, personal time-location-duration trajectories for large cohorts

69

or on a population scale are generally missing. Therefore, in most epidemiological

70

studies only the mean concentration at the residential address is used to represent the

71

individual-specific exposure throughout the day 9. Recent studies proposed to account

72

for different microenvironments in which the subjects were present when estimating

73

exposure

7-12

. Few studies achieved this by directly modelling the subjects’ daily


74


Page 4 of 27

activity patterns 8 whereas other tracked the individuals’ mobility patterns via mobile

75

telephone signals (either using GPS or triangulation) or by simulating commuting

76

habits using, e.g., agent-based models 9, 11. In some cases, significant differences were

77

found between the estimated exposures at the residence place and at the locations

78

where the people were present according to the simulations

9, 12

. Moreover, although

79

pollutant concentrations are known to vary throughout the day due to changes in the

80

wind field, emission patterns, removal and fate processes, etc., the relative impact on

81

exposure misclassification of the subject’s mobility and of the temporal concentration

82

variation has been generally overlooked. Understanding the relative contributions of

83

these factors to exposure misclassification is important, since accounting for each of

84

them when estimating exposure may be time consuming and computationally

85

demanding. This work studies in fairly detail and using a large real cohort exposure

86

estimation differences that may occur when accounting for the persons’ mobility and

87

the pollutant concentration variability throughout the day. Due to lack of complete

88

information, as noted above, we tracked the cohort subjects’ daily trajectories while

89

accounting for a limited set of typical microenvironments in which people tend to

90

spend their time. The implication of our results for epidemiological studies is

91

discussed.

92 93

Methods

94

Study area and population

95

The study area covers Israel’s central coastal plain, from the city of Netanya in the

96

north to the city of Ashdod in the south (70 x 25 km), including the Tel-Aviv

97

metropolitan area. Geocoded home and work addresses of ~168,000 working adults

98

(age 24-65) that live and work in the study area were obtained from the Israeli Central

99


Page 5 of 27


Bureau of Statistics (CBS). This database is a representative sample (both spatially

100

and ethnically) of all the Israeli adult workers in this age group, and was generated by

101

the Israel Central Bureau of Statistics (ICBS) based on the 2008 census. Due to

102

privacy issues, the geocoding of the addresses was to the coordinates of the centers of

103

the census tracts in which the home or workplace of each individual were located. To

104

reduce the inherent inaccuracy of such a data aggregation, the centers of the census

105

tracts were calculated while accounting only for the built area in each tract. Moreover,

106

for subjects that reside or work in very dense urban areas, the coordinates of the

107

centers of the buildings’ rooftops (home and workplace) were available.

108 109

Pollutant concentrations

110

The dynamic Optimized Dispersion Model (ODM)

13

was used for calculating

111

ambient nitrogen dioxide concentrations in the study area. The ODM accounts for

112

temporal variation of the wind field at a half hourly resolution, and uses as a proxy of

113

the spatiotemporal distribution of pollutant emissions traffic volumes in ~11,500 road

114

network segments, obtained using an operative traffic assignment model and casted

115

into structured grid (500 x 500 m). For every half-hour, the model optimizes the

116

parameters of a nonlinear simple yet physically sound dispersion scheme, using the

117

half-hourly monitoring records from 25 population monitoring stations as the

118

dependent variables. The model then projects the concentrations to all the grid cells.

119

The model was used for calculating half-hourly NO2 concentrations for the whole

120

year of 2008, with its performance thoroughly evaluated

13

. In this work, the half-

121

hourly concentrations at each grid cell were averaged over five time windows of

122

typical activity periods, described in the next section, to produce a set of five

123

concentration values assigned to each of the grid cells. The concentrations which we

124



Page 6 of 27

used, and the exposure derived from them for each subject based on the individual-

125

specific simulated daily activity patterns, represent a typical day in 2008.

126 127

Integrated daily exposure

128

In general, personal exposure can be calculated as the integration of the pollutant

129

concentrations to which an individual is exposed over contact-time increments. In

130

practice, the exposure is usually estimated as a summation of the average

131

concentrations that a subject encounters while spending time (over which the

132

averaging is performed) at certain distinct microenvironments visited throughout the

133

daily routine,

134

E= ∙ ≈ ∑ ∙ = ∙ ̅ ,

(1)

135

where E is the personal exposure to a given pollutant, ci is the average pollutant

136

concentration in the ith microenvironment during the time the individual is present in

137

the microenvironment, ti is the time spent by the individual in the ith

138

microenvironment, T is the total exposure time, e.g. 24 h when the exposure is

139

assessed over the entire day, and ̅ is the time-location-duration weighted average

140

concentration to which the individual has been exposed during time-period T.

141

Naturally, dividing Eq. (1) by T results in the daily average concentration, weighted

142

according to the time spent in each microenvironment. The daily average pollutant

143

concentration is oftentimes referred to as the daily exposure, with the common

144

understanding that the actual exposure (in terms of dose) is the product of ̅ times T.

145

For both simplicity and generalizability, we accounted for five typical daily activity-

146

periods: at home, morning commute to work, at work, evening commute back home,

147

and leisure. These activities are uniquely linked to the individual-specific location and

148

time of the day, i.e. where, when and for how long these activities take place. It is

149


Page 7 of 27


noteworthy that since concentrations inside the homes, workplaces or transportation

150

microenvironments were unavailable to us, the exposure of each individual is

151

estimated based on the outdoor concentrations at these locations, similarly to the

152

common practice of taking the outdoor residential concentration as a proxy of

153

exposure.

154

For simplicity, we consider only working days. Clearly, a true long-term

155

exposure estimation should account also for the time-location-duration activities and

156

concentration levels during weekends. Yet, this requires individual-level trajectory

157

data and preferences that we did not have. The individual-specific workday daily

158

trajectory was estimated based on general behavioral patterns in Israel (which may be

159

relevant also for other places). All the study subjects were assumed to be at home for

160

10 hours, from 21:00 until 07:00, and at work for 9 hours, from about 08:00 until

161

about 17:00. The time at which each individual starts working varies slightly,

162

depending on the individuals’ commute. As an initial guess, the morning commute

163

was assumed to start at 07:00 and the evening commute was assumed start at 17:00,

164

with both periods lasting for one hour. The leisure time completes the individual daily

165

activity pattern, and was initially assumed to extend from 18:00 to 21:00. The

166

individual-specific commute and leisure times were then modified as follows. An ad-

167

hoc Python code was used to find the shortest route that connects the individual-

168

specific home and work addresses (in practice, the centroids of the home and work

169

census tracts), using the network analysis layer of Israel and the Network Analyst

170

extension of ArcGIS 10.1 (ESRI, USA). The code outputs the shortest route for each

171

person, accounting for the driving direction in one way roads and the average traffic

172

conditions. The mean speed at each hour of the day along each segment of the road

173

network has been obtained from on-board vehicle’s GPS signals, collected and

174



Page 8 of 27

processed by Decell Ltd, Tel Aviv, Israel. Consequently, the morning and evening

175

commuting times of each individual were calculated based on the length and the mean

176

driving speed at each road segment along the individual’s commuting route. Each

177

route was divided into segments and the route mean traveling velocity or the

178

concentrations were assigned to each segment vertices. The average pollutant

179

concentration during the commute was calculated as the weighted average

180

concentration, or velocity, in each road segment along the route during the commuting

181

time.

182

Since specific information on the leisure habits of the Israeli adult population is

183

not available, the venue and duration of leisure had to be estimated for each

184

individual. Based on subjective judgment, we assumed that Israeli adults normally

185

spend their leisure time in areas close to their residences, either in commercial centers

186

or in designated areas of cultural, recreational or sport activities. These types of

187

leisure-activity areas were located in a GIS layer of Israel's land use map, and the

188

nearest area within a distance of 3 km to the residence place of each individual was

189

considered as the leisure time venue. For each individual, the leisure time was set

190

such that it completes the 24 h daily cycle, with the time spent at home and at work

191

fixed. Thus, when the total daily commute (home to work and back) was less (more)

192

than two hours we added (subtracted) the extra time to (from) the daily leisure time.

193 194

Exposure misclassification

195

To assess the possible magnitude of exposure differences arising due to accounting

196

for the daily mobility, temporal concentrations variability, or both, we compared four

197

different exposure scenarios. In scenario (1), our benchmark, the 24 h exposure is

198

estimated as the equally-weighted 24 h mean ambient pollutant concentrations at the

199


Page 9 of 27


subject’s home address. This scenario does not account for the subject’s mobility

200

(designated here as static mobility, SM), and the concentration to which the subject is

201

exposed is the daily average concentration (designated here static concentration, SC).

202

Hence, this scenario is termed SMSC. The three other scenarios strive to better

203

approximate the exposure E (Eq. 1) by considering mobility of the study subjects

204

throughout the day in between the five

generic microenvironments (designated

205

dynamic mobility, DM), or/and considering a dynamic concentration variability

206

throughout the day (designated dynamic concentrations, DC). In particular, the

207

alternative scenarios are: (2) static mobility and dynamic concentrations (SMDC), (3)

208

dynamic mobility and static concentration (DMSC), and (4) dynamic mobility and

209

dynamic concentrations (DMDC). Comparisons between the four scenarios are used

210

for elucidating the effect of accounting for the individuals’ mobility and for the

211

dynamic pollutant concentration field on exposure misclassification.

212

The differences in exposure estimates based on the three more complex

213

scenarios relative to the benchmark scenario are calculated for each person separately

214

and presented as a frequency distribution. After normalization, the distributions

215

represent the probability density functions (PDF) of the exposure estimation

216

differences. To obtain a better insight into the specific contributions to the total

217

differences between the exposure scenarios, we examined for each individual also the

218

differences per microenvironment. Statistics used for studying the exposure

219

differences include the root mean squared error (RMSE) and the coefficient of

220

determination (R2).

221 222

Sensitivity analysis

223



Page 10 of 27

A major shortcoming of the methodology we used to calculate the integrated exposure

224

is the lack of actual information on the time spent by each individual in every

225

microenvironment. To assess the uncertainty associated with our rather arbitrary daily

226

trajectory assumptions (five general activities and durations), we carried out an in-

227

depth sensitivity analysis. A random sample of 10,000 individuals was selected out of

228

the total study population and for each of them we calculated the daily exposure based

229

on 17 different time-activity patterns. Each of these patterns is a variation of the basic

230

one (10 h at home and 9 h at work), with the time spent at home ranging from 8 to 16

231

h at a half-hourly resolution, and the time at work varying, correspondingly, between

232

11 and 3 h. Namely, in all these daily-activity patterns the time spent by any

233

individual at both home and work always sums up to 19 h. The commute time was set

234

based on the shortest route and average vehicle speed per road segment, as explained

235

above. Hence, the commute time as well as the leisure time were both assumed to be

236

independent of the sum of the two time periods spent at home and at work. The model

237

sensitivity to the uncertainty in the time-activity patterns is reported in terms of the

238

distribution of the coefficients of variation (CV) of the exposure estimates that were

239

calculated for the randomly selected 10,000 individuals, based on the 17 time-activity

240

patterns specified for each of them.

241 242

Results

243

Figure 1 depicts the sensitivity of the integrated daily exposure estimates to the time-

244

activity pattern, based on the DMDC (i.e. most complex) scenario. The range of CV

245

values (0-10%) suggests a relatively small sensitivity of the exposure estimates to the

246

rather large uncertainty in the individual’s daily time-location-duration trajectory. We

247

can thus expect that our exposure estimates under the DMSC scenario will also

248


Page 11 of 27


contain no more than 10% uncertainty due to the somewhat arbitrary choice of time

249

periods spent at home and at work.

250 251

252

Figure 1. Distribution of the coefficient of variation (CV) of the integrated daily

253

exposure of a random sample of 10,000 individuals based on the DMDC scenario,

254

with the individual-specific CV obtained by varying the home occupancy hours (at the

255

expense of the working hours away from home), as part of the sensitivity analysis

256

exploration (see Methods: Sensitivity analysis for more details).

257 258

Figure 2a demonstrates the distribution of NO2 concentrations in the various

259

microenvironments during the time-of-day spent in each of them by each subject. It is

260

evident that the differences between the mean concentrations among the different

261

microenvironments are small compared to the concentration variability within each of

262

them. The wide concentration distribution in each of the microenvironments suggests

263

that while the average concentrations in the different microenvironments are quite

264

similar, some subjects may experience large concentration variation when moving

265

between them. Figure 2b depicts the distributions of the exposures at the different

266

microenvironments based on the DMDC scenario. The contributions of the exposure

267

at each microenvironment (i.e. during different activities) to the integrated daily

268



Page 12 of 27

exposure can be clearly appreciated. Large and statistically significant (p

The impacts of personal mobility and diurnal ... - ACS Publications

Recommend Documents