Evaluating the Impact of Neighborhood Characteristics on

Subscriber access provided by Kaohsiung Medical University

Ecotoxicology and Human Environmental Health

Evaluating the Impact of Neighborhood Characteristics on Differences between Residential and Mobility-based Exposures to Outdoor Air Pollution Masoud Fallah Shorshani, Marianne Hatzopoulou, Nancy Ross, Zachary Patterson, and Scott Weichenthal Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.8b02260 • Publication Date (Web): 17 Aug 2018 Downloaded from http://pubs.acs.org on August 20, 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.

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 32

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Environmental Science & Technology

Evaluating the Impact of Neighborhood Characteristics on Differences between Residential and Mobility-based Exposures to Outdoor Air Pollution Masoud Fallah-Shorshani1, Marianne Hatzopoulou2, Nancy A. Ross3, Zachary Patterson4, Scott Weichenthal*1 1

McGill University, Department of Epidemiology, Biostatistics and Occupational Health,

15

Montreal, Quebec H3A 1A2, Canada

16 17 18 19

2

University of Toronto, Department of Civil Engineering, Toronto, Ontario, M5S 1A4, Canada

20

4

21

Quebec, HG3 1M8

3

McGill University, Department of Geography, Montreal, Quebec H3A 2K6, Canada Concordia University, Department of Geography, Planning and Environment, Montreal,

22 23

*Corresponding Author

24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

Scott Weichenthal Faculty of Medicine Department of Epidemiology, Biostatistics, and Occupational Health McGill University 1020 Pins Ave. West Montreal, QC H3A 1A2, Canada Email: [email protected] Tel: (514) 398-1584

Abstract

1 ACS Paragon Plus Environment


40

Epidemiological studies often assign outdoor air pollution concentrations to

41

residential locations without accounting for mobility patterns. In this study, we examined

42

how neighborhood characteristics may influence differences in exposure assessments

43

between outdoor residential concentrations and mobility-based exposures. To do this,

44

we linked residential location and mobility data to exposure surfaces for NO2, PM2.5, and

45

ultrafine particles in Montreal, Canada for 5452 people in 2016. Mobility data were

46

collected using the MTL Trajet smartphone application (mean: 16 days/subject).

47

Generalized additive models were used to identify important neighborhood predictors of

48

differences between residential and mobility-based exposures and included residential

49

distances to highways, traffic counts within 500 meters of the residence, neighborhood

50

walkability, median income, and unemployment rate. Final models including these

51

parameters provided unbiased estimates of differences between residential and

52

mobility-based exposures with small root mean square error values in 10-fold cross

53

validation samples. In general, our findings suggest that differences between

54

residential and mobility-based exposures are not evenly distributed across cities and

55

are greater for pollutants with higher spatial variability like NO2. It may be possible to

56

use neighborhood characteristics to predict the magnitude and direction of this error to

57

better understand its likely impact on risk estimates in epidemiological analyses.

58 59 60 61 62 63 64 65 66

1.

Introduction


Page 2 of 32

Page 3 of 32


Large population-based studies of the long-term health effects of outdoor air

67 68

pollution generally rely on exposure estimates assigned to residential locations.1-2 This

69

is a reasonable approach given other options are not readily available, and people often

70

spend a large portion of their time in and around their home. This approach, however,

71

has clear limitations as it does not account for individual-level mobility patterns, which

72

may act to increase or decrease long-term exposures relative to outdoor residential

73

concentrations. Moreover, it is not clear how urban environments (e.g. local traffic

74

density, access to transit, walkability) may impact differences between residential and

75

mobility-based-exposure measures, or how these patterns may differ between

76

pollutants. These are important considerations as the validity of residential outdoor

77

concentrations as surrogate measures of long-term personal exposures to outdoor air

78

pollution has a direct impact on the validity of health risk estimates derived from these

79

exposures.

80

Previous studies have reported important differences between personal

81

exposures to air pollution and those estimated using mobility information, home/work

82

locations, or commute distance.3-5 In addition, studies based on mobility surveys6

83

suggest that the health risks of air pollution may be underestimated when mobility is not

84

considered in the exposure assessment process and that this bias may be stronger

85

when the spatial variability of pollution concentrations is greater. Other studies have

86

used activity space questionnaires to analyze the impact of non-residential

87

neighborhood exposures on self-reported health7,8 and Chaix et al.9 proposed an

88

integrated approach combining GPS tracking, accelerometers, and an electronic web-

89

based mobility survey to improve measures of exposure by accounting daily mobility



90

patterns. Smartphone applications have also been developed to track time-activity

91

patterns for the purpose of exposure assessment10,11 but to our knowledge studies have

92

yet to evaluate important neighborhood-level predictors of differences (i.e. potential

93

bias) between mobility-based exposures and residential exposure estimates. Here we address this question by comparing residential PM2.5, NO2, and ultrafine

94 95

particle (UFPs, 55 years) and were predominantly male (61%). On average, residential and mobility-based exposure estimates were similar for

207 208

the population as a whole (Figure 2 and Supplemental Material: Table S1). However,

209

overall averages masked important spatial patterns that were apparent when we

210

mapped differences between residential and mobility-based exposures across

211

dissemination areas in Montreal. Specifically, Figure 3 was generated by first calculating

212

individual differences between residential and mobility-based exposures for each

213

subject in our study, assigning these differences to residential locations, and then

214

averaging these differences for all residential locations within each dissemination area

215

(areas with no data are shown as white in Figure 3). In general, residential exposure

216

estimates for NO2, PM2.5, and UFPs in the suburban areas on the west portion of the

217

island tended to underestimate mobility-based exposures (residential exposure is

218

smaller than mobility exposure), whereas residential estimates in the downtown core

219

tended to overestimate mobility-based exposures. Scatter plots of residential

220

concentrations versus differences between mobility-based and residential estimates are

221

shown in Supplemental Figure S1. These plots illustrate a trend whereby individual-level

222

mobility patterns tend to increase exposures for people living in low exposure areas and

223

decrease exposure for people living in high exposure areas. Simple scatter plots of

224

mobility-based versus residential exposure estimates are shown in Supplemental Figure

225

S2. Descriptive data for the built environment and land use parameters examined in

226 227

the GAM models are listed in Table 1; variables retained in final models are listed in



228

Table 2. Most parameters were not strongly correlated (r < 0.55) with the exception of

229

length of bike lanes and length of roads (r=0.95). Length of bike lanes was removed

230

from the UFP model as this was the only model that included both of these parameters.

231

The relationships between various built-environment/neighborhood characteristics and

232

differences between residential and mobility-based exposure are illustrated in Figures 4-

233

6. Many of these relationships were non-linear and the magnitudes of differences

234

across the various built-environment/neighborhood factors were greater for NO2 and

235

UFPs than for PM2.5. For NO2 (Figure 4), residential exposures tended to underestimate mobility-

236 237

based exposures with increasing distance from metro lines (panel a) and as median

238

income (panel f) and park space (panel i) increased. On the other hand, residential

239

exposures tended to overestimate mobility-based estimates as dissemination area

240

unemployment rates increased (panel j) and as the length of highways within 100 m of

241

the residence increased (panel h). Residential exposures tended to underestimate

242

mobility-based exposures for subjects living in the least walkable areas and

243

overestimate exposures in the most walkable areas (panel k). In some cases, u-shaped

244

(distance to highways, length of bus lanes) or inverted u-shaped (distance to

245

expressway) relationships were observed with the magnitude of over/underestimation

246

varying across the distribution of land use factors. For UFPs, residential exposures tended to overestimate mobility-based

247 248

estimates as the length of bus lanes, highways, and rail within 100 m increased around

249

residences (panels b, c, and f). An inverted u-shaped relationship was observed for the

250

length of streets within 100 m buffers of residences with residential concentrations


Page 10 of 32

Page 11 of 32


251

overestimating mobility-based exposures as the length of streets within this buffer

252

increased (panel h). As for NO2, residential UFP exposures tended to underestimate

253

mobility-based estimates for subjects living in the least walkable areas and overestimate

254

exposures in the most walkable areas (panel g).

255

For PM2.5, residential exposures tended to underestimate mobility-based

256

exposures as distance to metro lines, traffic counts (100 m buffer), and length of bus

257

lanes (100 m buffer) increased (panels a, b, and k). Inverted u-shaped relationships

258

were observed for distance to expressways, park space (100 m buffer), and length of

259

railways (100 m buffer) with residential exposures tending to overestimate mobility-

260

based measures at higher values of these parameters (panels e, d, and l). Finally, u-

261

shaped relationships were observed for walkability, median income, unemployment

262

rate, and length of bike lanes (100-m buffer) (panels j, n, o, and m). As for UFPs and

263

NO2, residential PM2.5 exposures tended to underestimate mobility-based estimates for

264

subjects in the least walkable areas. The results of model evaluations using 10-fold cross-validation procedures

265 266

(n=545 per iteration) are presented in Table 3. All three models provided unbiased

267

estimates of differences between outdoor residential concentrations and mobility-based

268

air pollution exposures. R2 values were low but RMSE values were small relative to the

269

typical exposures levels (12%, 16%, and 3.0% of residential concentrations for NO2,

270

UFPs, and PM2.5, respectively) suggesting that these models may still be useful in

271

practice given that they also provide unbiased estimates. Indeed, the low R2 values are

272

likely due in part to the narrow range of differences modeled and, in this case, RMSE

273

values likely provides a more useful measure of model utility.8



274 275 276

4.

Discussion

277

evaluate the long-term health effects of these exposures. This is a reasonable

278

approach, but one clear limitation is their inability to capture exposure variations

279

resulting from individual-level mobility patterns. This is an important issue as the validity

280

of risk estimates derived from these studies is directly linked to the extent to which

281

residential concentrations represent long-term personal exposures to air pollution of

282

ambient origin.

Residential estimates of outdoor air pollution concentrations are widely used to

283

We examined how the urban built environment and various neighborhood

284

characteristics may influence differences between residential and mobility-based

285

exposures using individual-level mobility data and exposure models for outdoor NO2,

286

UFPs, and PM2.5. Our results suggest that there may be substantial within-city spatial

287

variations in the validity of outdoor residential concentrations as estimates of long-term

288

exposures to outdoor air pollution. The magnitude of the difference between residential

289

and mobility-based estimates appears to be greater for pollutants with more spatial

290

variability (e.g., NO2 and UFP) and is influenced by a number of built

291

environment/neighborhood characteristics including residential proximity to traffic,

292

neighborhood walkability, median income, and unemployment rate. Importantly, our

293

results suggest that it may be possible to use various built environment/neighborhood-

294

level characteristics to adjust residential concentrations to more accurately reflect

295

mobility-based exposures. Additionally, future studies could use this approach to

296

estimate the magnitude and direction of this error to better understand the likely impact

297

on risk estimates in epidemiological analyses. However, the additional effort required to


Page 12 of 32

Page 13 of 32


298

implement this analysis may not be equally justifiable for all pollutants. Specifically, our

299

findings suggest that differences between residential and mobility-based exposures are

300

greatest for pollutants with high spatial variability (i.e. UFPs and NO2) with smaller

301

differences observed for PM2.5. Therefore, it may be more efficient to focus on other

302

sources of error (e.g. penetration of outdoor air indoors) for pollutants with lower spatial

303

variations.

304

To our knowledge this is the first study to evaluate how various built environment

305

and neighborhood level characteristics may impact differences between residential and

306

mobility-based exposures to outdoor air pollution using individual-level mobility data.

307

Tonne et al.27 noted that residential exposures consistently overestimated personal

308

exposures in London, UK based on simulated routes between origins and destinations

309

identified in London’s annual travel demand survey. This team also reported that

310

overestimation depended in part on household income and area-level income

311

deprivation and our findings for median income and unemployment rate in Montreal are

312

consistent with these results. In general, predictors of potential bias in residential

313

exposure estimates may be city-specific owing to different spatial patterns of

314

populations in relation to various built environment/neighborhood-level factors.

315

However, some patterns may be more generalizable across cities including those for

316

suburban commuting to downtown (i.e. mobility-bases exposures are likely higher than

317

residential concentrations). Our results and those of Tonne et al.27 do, however, point to

318

several issues that should be considered in future epidemiological investigations. First,

319

the magnitude (and direction) of exposure measurement error is likely not constant for

320

populations distributed across cities when residential outdoor air pollution



321

concentrations are used as the primary exposure variable. The impact of this error on

322

risk estimates will depend on the measurement error structure in any given study and

323

could result in bias either toward or away from the null. In the future, more attention

324

should be paid to the impact that these various built environment/neighborhood-level

325

factors could have on the magnitude of exposure measurement error in a given study in

326

order to minimize their impact on study results or at least understand the likely direction

327

of potential bias in risk estimates. While this study had a number of methodological strengths including a detailed

328 329

time-series of individual-level mobility data mapped to multiple air pollution exposure

330

surfaces, we note several limitations. First, our study population is not representative of

331

the overall population of Montreal and in particular our sample included few elderly

332

people. As a result, our results may not reflect the magnitudes/directions of potential

333

differences between residential and mobility-based exposures for subjects outside the

334

age range included in our study. Similarly, participant residential locations were not

335

randomized and thus our sample may not be representative of spatial patterns across

336

the entire island of Montreal (although we did have participants living in most areas of

337

the island). In addition, the spatial resolution of our PM2.5 exposure surface was limited

338

to 1 km2 and our ability to identify differences between residential and mobility-based

339

exposure estimates was less than for NO2 or UFPs. In practice, however, within-city

340

spatial differences in outdoor PM2.5 concentrations vary less than these pollutants and

341

our results would not be expected to change dramatically if a higher resolution model

342

were available.


Page 14 of 32

Page 15 of 32


We did not attempt to evaluate personal air pollution exposures in homes or in

343 344

transit environments (e.g., subways, buses). Instead, our comparisons focused on

345

residential outdoor concentrations and outdoor concentrations encountered along

346

typical outdoor daily mobility paths. Some may view this as a limitation, but transit-

347

specific exposures are not typically considered in epidemiological studies of long-term

348

exposure to outdoor air pollution as the parameters needed to estimate these values

349

are not readily available for application to large population-based cohorts. Likewise, we

350

did not consider built environment factors encountered along mobility paths as

351

predictors in our models as this information would also not be available in population-

352

based studies. Moreover, our database did not support a detailed evaluation of

353

individual-level factors that may influence differences between residential and mobility-

354

based exposures. Future studies should explore these important questions. Finally,

355

while the application used in this study (MTL Trajet) provided an efficient means of

356

collecting mobility data, the block-level spatial accuracy of this application may have

357

underestimated spatial variations in mobility-based exposures and future studies will

358

aim to refine this metric.



In summary, residential estimates of long-term exposures to outdoor air pollution

359 360

may not adequately reflect exposures that account for individual-level mobility patterns.

361

This is particularly true for pollutants with high spatial variability like UFPs and NO2 and

362

may be less of a concern for pollutants with less spatial variability such as PM2.5. In

363

some cases, it may be possible to use various built environment and neighborhood level

364

characteristics to estimate differences between residential concentrations and mobility-

365

based exposures which in turn may help to evaluate the likely impact of this source of

366

measurement error on health risk estimates.

367 368

Supporting Information

369

Supplemental Methods: The MTL Trajet Study; Descriptive statistics for residential and

370

mobility-based exposures (Table S1); Scatter plots of differences between mobility and

371

residential exposures versus residential estimates for UFPs, PM2.5, and NO2 (Figure

372

S1); Scatter plots of mobility-based versus residential exposure estimates for UFPs,

373

PM2.5, and NO2 (Figure S2).

374 375 376 377 378 379 380 381 382 16 ACS Paragon Plus Environment

Page 16 of 32

Page 17 of 32


383 384

References

385

(1)

Pinault, L.L.; Weichenthal, S.; Crouse, D.L.; Brauer, M.; Erickson, A.; Donkelaar,

386

A.V.; Martin, R.V.; Hystad, P.; Chen, H.; Fines, P.; Brook, J.R.; Tjepkema, M.;

387

Burnett, R.T. Associations between fine particulate air pollution and mortality in

388

the 2001 Canadian Census Health and Environment Cohort. Environ. Res. 2017,

389

159, 406-415.

390

(2)

Turner, M.C.; Jerrett, M.; Pope, C.A. 3rd; Krewski, D.; Gapstur, S.M.; Diver, W.R.;

391

Beckerman, B.S.; Marshall, J.D.; Su, J.; Crouse, D.L.; Burnett, R.T. Long-term

392

ozone exposure and mortality in a large prospective study. Am. J. Respir. Crit.

393

Care Med. 2016, 193 (10), 1134-1142.

394

(3)

Quackenboss, J.J.; Spengler, J.D.; Kanarek, M.S.; Letz, R.; Duffy, C.P. Personal

395

exposure to nitrogen-dioxide — relationship to indoor outdoor air-quality and

396

activity patterns. Environ. Sci. Technol. 1986, 20 (8), 775–783.

397

(4)

Kousa, A.; Monn, C.; Rotko, T.; Alm, S.; Oglesby, L.; Jantunen, M.J. Personal

398

exposures to NO2 in the EXPOLIS-study: relation to residential indoor, outdoor

399

and workplace concentrations in Basel, Helsinki and Prague. Atmos. Environ.

400

2001, 35(20), 3405–3412.

401

(5)

Nethery, E.; Leckie, S.E.; Teschke, K.; Brauer, M. From measures to models: an

402

evaluation of air pollution exposure assessment for epidemiological studies of

403

pregnant women. Occup. Environ. Med. 2008, 65(9), 579–586.

404

(6)

Setton, E.; Marshall, JD.; Brauer, M.; Lundquist, KR.; Hystad, P.; Keller, P.; Cloutier-Fisher D. The impact of daily mobility on exposure to traffic-related air

405



406

pollution and health effect estimates. J. Expo. Sci. Environ. Epidemiol. 2011,

407

21(1), 42-48.

408

(7)

Inagami, S.; Cohen, DA.; Finch, BK. Non-residential neighborhood exposures

409

suppress neighborhood effects on self-rated health. Soc. Sci. Med. 2007, 65(8),

410

1779-1791.

411

(8)

Vallée, J.; Cadot, E.; Grillo, F.; Parizot, I.; Chauvin, P. The combined effects of

412

activity space and neighbourhood of residence on participation in preventive

413

health-care activities: The case of cervical screening in the Paris metropolitan

414

area (France). Health Place 2010, 16(5), 838-852.

415

(9)

Chaix, B.; Meline, J.; Duncan, S.; Merrien, C.; Karusisi, N.; Perchoux, C.; Lewin,

416

A.; Labadi, K.; Kestens, Y. GPS tracking in neighborhood and health studies: a

417

step forward for environmental exposure assessment, a step backward for causal

418

inference? Health Place 2013, 21: 46-51.

419

(10)

De Nazelle, A.; Seto, E.; Donaire-Gonzalez, D.; Mendez, M.; Matamala, J.;

420

Nieuwenhuijsen, MJ.; Jerrett, M. Improving estimates of air pollution exposure

421

through ubiquitous sensing technologies. Environ. Pollut. 2013, 176, 92-99.

422

(11)

Glasgow, ML.; Rudra, CB.; Yoo, EH.; Demirbas, M.; Merriman, J.; Nayak, P.;

423

Crabtree-Ide, C.; Szpiro, AA.; Rudra, A.; Wactawski-Wende, J.; Mu, L. Using

424

smartphones to collect time–activity data for long-term personal-level air pollution

425

exposure assessment. J. Expo. Sci. Environ. Epidemiol. 2016, 26(4), 356-364.

426

(12)

Patterson. Z. MTL Trajet 2016. Paper presented at the 11th International

427

Conference on Travel Survey Methods, Esterel, Quebec 2017, Available at:

428

itinerum.ca/documents.html


Page 18 of 32

Page 19 of 32

429


(13)

https://ville.montreal.qc.ca/mtltrajet/en/

430 431

Ville de Montreal; MTL Trajet 2017, Accessed on June 15, 2017,

(14)

Technical report Concordia University TRIP Lab 2017. The Itinerum Open

432

Smartphone Travel Survey Platform., available at:

433

www.itinerum.ca/documents.html

434

(15)

J. Trans. Res. Board 2016, 2594, 35-43.

435 436

Patterson, Z.; Fitzsimons, K. DataMobile: smartphone travel survey experiment.

(16)

The Communauté Métropolitaine de Montréal - CMM

437

http://cmm.qc.ca/fileadmin/user_upload/documents/20180213_CMMEnChiffres.p

438

df

439

(17)

the Canadian Census Analyzer: 2017,  http://dc1.chass.utoronto.ca/census,

440 441

Statistics Canada, 2017. Profile of census tracts of the census. accessed through

(18)

Weichenthal, S.; Ryswyk, K.V.; Goldstein, A.; Bagg, S.; Shekkarizfard, M.;

442

Hatzopoulou, M. A land use regression model for ambient ultrafine particles in

443

Montreal, Canada: a comparison of linear regression and a machine learning

444

approach. Environ. Res. 2016, 146: 65-72.

445

(19)

Deville Cavellin, L.; Weichenthal, S.; Tack, R.; Ragettli, M.S.; Smargiassi, A.;

446

Hatzopoulou, M. Investigating the use of portable air pollution sensors to capture

447

the spatial variability of traffic-related air pollution. Environ. Sci. Technol. 2016,

448

50 (1), 313-320.

449

(20)

Van Donkelaar, A.; Martin, R.V.; Spurr, R.J.D.; Burnett, RT. High-resolution satellite-derived PM2.5 from optimal estimation and geographically weighted

450



451

regression over North America. Environ. Sci. Technol. 2015, 49 (17), 10482-

452

10491.

453

(21)

DMTI: DMTI Spatial Inc. Database 2009. CanMap Street files. 2010.

454

(22)

Sider, T.; Alam, A.; Zukari, M.; Dugum, H.; Goldstein, N.; Eluru, N.; Hatzopoulou,

455

M. Land-use and socio-economics as determinants of traffic emissions and

456

individual exposure to air pollution. J. Transp. Geogr. 2013, 33, 230-239.

457

(23)

PTV Vision. (2009). VISUM 11.0 Basics. Karlsruhe, Germany: PTV AG.

458

(24)

Canadian Active Living Environments Database (Can-ALE) User Manual &

459

Technical Document (http://canue.ca/wp-

460

content/uploads/2018/03/CanALE_UserGuide.pdf)

461

(25)

Routledge. 2017, 249-307.

462 463

Hastie, Trevor J. "Generalized additive models." Statistical models in S.

(26)

Alexander, D.L.J; Tropsha, A.; Winkler, A. Beware of R2: simple, unambiguous

464

assessment of the prediction accuracy of QSAR and QSPR Models. J. Chem.

465

Inf. Model. 2015, 55 (7), 1316-1322.

466

(27)

Tonne, C.; Mila, C.; Fecht, D.; Alvarez, M.; Gulliver, J.; Smith, J.; Beevers, S.;

467

Anderson, HR.; Kelly, F. Socioeconomic and ethnic inequalities in exposure to air

468

and noise pollution in London. Environ. Int. 2018, 115, 170-179.

469 470 471 472 473


Page 20 of 32

Page 21 of 32


474 475 476 477 478 479 480

TABLES

481

Table 1. Descriptive statistics for land use and built environment parameters Variable

Best buffer

Mean

Median

Maximum

Minimum

Length Variables (m) Length of bus lanes

100

353

241

4350

0

Length of the bike lanes

100

539

539

1745

0

Length of the highways

100

22

0

1415

0

Length of the all roads

100

545

542

1745

197

Length of rail lines

100

2

0

590

0

Distance from metro line

1766

737

24,070

1

Distance from expressways

1552

1450

5753

5

Distance from highways

733

564

4826

1

Distance from the shore

1261

1129

4004

25

500

7843

7125

32,662

194

Area of the open area

500

35,190

14,580

660,400

0

Area of the parks

500

55,330

38,030

492,800

0

Area of the governmental

500

66,630

54,590

719,400

0

Area of the industrial

500

118,600

88,200

712,100

0

9

8

29

0

Median Household Income ($)

41290

37264

210,809

17

Population Count

7471

7241

73,982

0

234,414

176,138

27,084,300

36,121

1.61

1.43

13.7

-8.49

Distance Variables (m)

Traffic Counts 2

Land Use Area (m )

Dissemination Area Variables Unemployment rate (%)

Real estate value ($) Walkability index

482 21 ACS Paragon Plus Environment


Page 22 of 32

483 484 485 486 487 488 489 490 491

Table 2. Final model parameters in GAM models for the difference between residential and mobility-based exposures to NO2, UFPs, and PM2.5 in Montreal, Canada NO2 Distance to metro line Distance to highway Traffic Counts Length of bus lines Walkability index Median income Unemployment rate Distance to expressway Distance to shore Parks Length of highway

2

492 493 494

UFPs Distance to metro line Distance to highway Traffic Counts Length of bus lines Walkability index Median income Unemployment rate Distance to shore Length of highway Length of all roads Length of railways Open area

2

PM2.5 Distance to metro line Distance to highway Traffic Counts Length of bus lines Walkability index Median income Unemployment rate Distance to expressway Length of bike lanes Length of railways Industrial area Open area Population Density Parks Age 2

LOOCV R = 0.30 LOOCV R = 0.14 LOOCV R = 0.13 LOOCV, leave one out cross validation results for model selection procedure

495 496 497 498 499 500 501


Page 23 of 32


502 503 504 505 506 507 508 509

Table 3. Model evaluation using a 10-fold cross-validation procedure for the difference between residential and mobility-based exposure to NO2, UFPs, and PM2.5 Linear Models for Measured vs. Predicted Values Pollutant NO2 (ppb) UFPs 3 (count/cm )

Mean Slope (95% CI) 1.00 (0.939, 1.06)

Mean R (95% CI) 0.29 (0.26, 0.32)

Mean RMSE (95% CI) 2.91 (2.81, 3.01)

195 (-64, 455)

0.93 (0.78, 1.08)

0.14 (0.11, 0.17)

4030 (3688, 4372)

3

-0.0111 0.96 0.13 0.26 (-0.0277, 0.00542) (0.81, 1.11) (0.089, 0.16) (0.25, 0.28) Results are based on 10 repeated hold out samples with the model developed on a random 90% of the data and tested on the remaining 10% (n=545) PM2.5 (µg/m )

510 511 512

2

Mean Intercept (95% CI) 0.0291 (-0.0495, 0.108)

513 514 515 516 517 518 519 520 521 522 523 524 23 ACS Paragon Plus Environment


525 526 527 528

FIGURES

529

Figure 1. Spatial distribution of study participants by 3-digit postal codes in Montreal,

530

Canada.

531 532 533 534 535 536 537 24 ACS Paragon Plus Environment

Page 24 of 32

Page 25 of 32

538 539 540 541 542 543 544 545 546 547 548


Figure 2. Distributions of residential and mobility-based estimates of ultrafine particle (UFP) (A), nitrogen dioxide (NO2) (B), and PM2.5 (C) exposures.

549 550 551 25 ACS Paragon Plus Environment


552 553 554 555 556 557 558 559 560 561 562 563

Figure 3. Spatial distributions of differences between residential and mobility-based exposure estimates for NO2 (A), UFPs (B) and PM2.5 (C) in Montreal, Canada. Red areas indicate dissemination areas where on average residential exposure estimates are greater than mobility-based exposures. Green areas indicate dissemination areas where on average residential exposure estimates are lower than mobility-based exposures. Dissemination areas with no subjects are colored white on the map. A

564 565

B


Page 26 of 32

Page 27 of 32

566 567


C

568 569 570 571 572 573 574 575 576 27 ACS Paragon Plus Environment


577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599

Figure 4. Relationships between built environment factors and differences between residential and mobility-based NO2 (ppb) exposures in Montreal, Canada. The red horizontal line indicates no difference between mobility-based exposures and residential estimates. Portions of the graphs above the red lines indicate when mobility-based exposures exceed residential estimates; portions below the red lines indicate when mobility-based exposures are lower than residential estimates.


Page 28 of 32

Page 29 of 32


a

600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621

b

c

d

e

f

g

h

i

j

k

Figure 5. Relationships between built environment factors and differences between residential and mobility-based UFP (per cm3) exposures in Montreal, Canada. The red horizontal line indicates no difference between mobility-based exposures and residential estimates. Portions of the graphs above the red lines indicate when mobility-based exposures exceed residential estimates; portions below the red lines indicate when mobility-based exposures are lower than residential estimates.



622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637

Page 30 of 32

a

b

c

d

e

f

g

h

i

j

k

l

Figure 6. Relationships between built environment factors and differences between residential and mobility-based PM2.5 (µg/m3) exposure in Montreal, Canada. The red horizontal line indicates no difference between mobility-based exposures and residential estimates. Portions of the graphs above the red lines indicate when mobility-based exposures exceed residential estimates; portions below the red lines indicate when mobility-based exposures are lower than residential estimates.


Page 31 of 32


a

d

e

f

g

h

i

j

k

l

m

n

638 639 640 641 642 643 644 645

c

b

TOC ART


o


646


Page 32 of 32

Evaluating the Impact of Neighborhood Characteristics on

Recommend Documents