Embed Size (px)
Citation preview
Validating Novel Air Pollution Sensors to Improve Exposure Estimates for Epidemiological Analyses and Citizen Science
Michael Jerrett1, David Donaire-Gonzalez2, Olalekan Popoola3, Roderic Jones3, Ronald C. Cohen4, Estela Almanza5, Audrey de Nazelle6, Iq Mead7, Glòria Carrasco-Turigas2, Tom Cole-Hunter2, Margarita Triguero-Mas2, Edmund Seto8, Mark Nieuwenhuijsen2
Author Affiliation
1 Department of Environmental Health and Center for Occupational and Environmental Health, Fielding School of Public Health, University of California, Los Angeles, United States ([email protected])2Center for Research in Environmental Epidemiology (CREAL), 08003 Barcelona, Catalonia, Spain ([email protected]; [email protected]; [email protected]: [email protected]; [email protected]) 3Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, UK ([email protected]; [email protected])4Department of Chemistry, University of California, Berkeley, 419 Latimer Hall, Berkeley, California 94720-1460, United States ([email protected])5 Environmental Health Sciences, School of Public Health, University of California, Berkeley, 50 University Hall, Berkeley, California 94720-7360,United States ([email protected])6Centre for Environmental Policy, Imperial College London, SW7 1NA, UK ([email protected])7 Iq Mead, Department of Chemistry, University of Manchester, UK ([email protected])8 Department of Environmental and Occupational Health, University of Washington, Seattle, Washington 98195, United States ([email protected])
Corresponding Author: Michael Jerrett, PhD Professor and Chair Department of Environmental Health Sciences Director, Center for Occupational and Environmental Health Fielding School of Public Health University of California, Los Angeles Tel: 310.825.9037 Email: [email protected]
123456789
10111213141516171819202122232425262728293031323334353637383940
Abstract
Low cost, personal air pollution sensors may reduce exposure measurement errors in epidemiological investigations and contribute to citizen science initiatives. Here we assess the validity of a low cost personal air pollution sensor. Study participants were drawn from two ongoing epidemiological projects in Barcelona, Spain. Participants repeatedly wore the pollution sensor − which measured carbon monoxide (CO), nitric oxide (NO), and nitrogen dioxide (NO2). We also compared personal sensor measurements to those from more expensive instruments. Our personal sensors had morderate to high correlations with government monitors with averaging times of 1-hour and 30-minute epochs (r ~ 0.38-0.8) for NO and CO, but had low to moderate correlations with NO2 (~0.04-0.67). Correlations between the personal sensors and more expensive research instruments were higher than with the government monitors. The sensors were able to detect high and low air pollution levels in agreement with expectations (e.g., high levels on or near busy roadways and lower levels in background residential areas and parks). Our finding suggest that the low cost, personal sensors have potential to reduce exposure measurement error in epidemiological studies and provide valid data for citizen science studies.
Word count: 192
2
414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485
86
123
Introduction
Efforts to characterize air pollution exposure in epidemiological and public health studies
have typically estimated ambient air pollution levels, based on the nearest routine
monitor or a prediction model such as disperson or land use regression models.1 These
estimates are then usuallly assigned to an individual through their home address.
Although important health risks have been revealed, reliance on proxy methods may
impart large exposure-measurement error. Depending on the exposure-error type, health
effect estimates may be attenuated and biased toward a null result, obscuring the true
benefits of air pollution control measures.2 This is particularly important for pollutants
with high spatial variability, such as traffic-related air pollutants.3
Innovations in science and technology such as mobile, personalised sensing now provide
opportunities to overcome limitations that have led to exposure-measurement errors.
These innovations also provide opportunities to understand multiple exposures in time
and space and are now spurring fields known as “ubiquitous” and “participatory” sensing
that have substantial relevance to the future of environmental epidemiology in particular,
but more generally for public health protection.4
We define ubiquitous sensing as a network of sensors, such as a dense array of air
pollution monitors, that have wide spatial coverage and are embedded in urban areas.
Participatory sensing is defined as a means of obtaining detailed information on personal
and population exposures via citizens volunteering to carry sensors to supply this data (as
3
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
456
citizen scientists) – often in exchange for useful information that might allow them to
better understand and prevent harmful exposures they face.5 Such definitions have
invariably fuzzy boundaries, where an exposure information gained from participatory
sensing may be used in tandem with information from a ubiquitous network to develop
more precise estimates of exposure.6 Ubiquitous and participatory sensors can improve
air pollution exposure estimates in both epidemiological studies and empowerment
exercises where citizen scientists seek to understand how ambient exposures could be
affecting their health.7 Such improvements in exposure assessment may refine the
estimates in of health effects from air pollution or give citizens better information on the
health risk they face from ambient exposures. In both instances, better exposure
assessments from sensors could result in improved public health protection.
While this kind of sensing shows excellent promise, there have been few published
attempts to validate how well the sensors function when deployed on free-living human
participants. Recent studies have demonstrated the utility of having personal
measurements of exposure and location to assess air pollution exposures, but these efforts
have used expensive, commercially-available sensors that in most instances cannot be
deployed en mass in larger epidemiological studies because of relatively high cost
($2000-10,000 USD per unit).8 In this paper we report on a series of validation studies for
a novel, low-cost personal air pollution sensor (i.e., less than $600 USD per unit).
Methods
4
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
789
Sensor Design: The personal sensors used here were designed and built at Cambridge
University, UK (for short we call them “CamPerS” for Cambridge Personal Sensors).
The CamPerS were designed to be compact and lightweight and thus convenient for
participants to carry. Electrochemical sensors from Alphasense Ltd. (UK) were
incorporated for carbon monoxide (CO), nitric oxide (NO), and nitrogen dioxide (NO2)
along with a temperature sensor, a Global Positioning System (GPS) and General Packet
Radio Service (GPRS) transmittor. All of the sensors are mounted behind a metal mesh
opening at one end of the unit (Figure 1 illustrates the version used in this validation
study). The sensors weigh ~450g with the batteries and ~ 330g without batteries.
Earlier work by Mead et al.9 gives more details on sensor design and laboratory and field
performance.
Field Studies: Field deployments occurred in two ongoing case-crossover studies
undertaken in Barcelona, Spain: (1) Positive Health Effects on the Natural Outdoor
Environment in Typical Populations of different regions in Europe (PHENOTYPE), and
(2) Transportation Air pollution and Physical ActivitieS (TAPAS) II Experimental Study
Extension.
The PHENOTYPE study involved 26 adults with poor mental health who visited three
environments: green (i.e. natural park), blue (i.e. beachfront) and urban (i.e. mixed-use
neighborhood). Psycho-physiological measures were taken before, during (at 30 and 210
minutes), and after each visit. Study participants were asked to stay in each of the
environments behaving as they would normally in that environment (while avoiding
5
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
101112
swimming, vigorous physical activity and only eating or drinking what was provided).
Participants were repeatedly monitored for air pollution with CamPerS, geographic
location, and physical activity (see Nieuwenhuijsen, 2013, for more details).10
The TAPAS II study involved 30 healthy, non-smoking adults who rode stationary
bicycles or sat resting in two contrasting environments – a high traffic zone on a bridge
above a major highway with substantial automobile and truck traffic and a low traffic
environment in a park with few immediate emission sources. Physiological measures
were taken before and after riding or resting in each setting. Study participants were
allowed to go about their normal lives in the interval between the scripted exposures and
their follow-up physiological measurements six hours later.
In both studies, numerous other research-grade instruments measuring similar parameters
to the CamPers were arrayed in proximity to the study participants during scripted
exposures.
Field data were collected between September 2013 and February 2014 by trained
technicians. CamPerS measured NO2, NO and CO on 10-second intervals. Participants
wearing the CamPerS also carried a cellular phone with software for measuring
geographic location and physical activity assessment (see de Nazelle et al. 2013 and
Donaire et al. 2013 for details of this assessment).10,11
6
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
131415
Validation Protocol: With our validations we sought to determine how well the CamPerS
could replicate measurements taken by either ratified government monitors or more
expensive, larger research-grade instruments. We also sought to determine whether the
monitors could classify meaningful differences among ambient and indoor micro-
environments based on samples collected by our study participants. To conduct our
validation, we followed four steps:
1. We calibrated the CamPerS in chamber experiments to determine the zero value
for each sensor. This involved constructing a pollution chamber and filling it with
purified zero air and running controlled experiments. We also conducted bump
tests where higher levels were introduced into the chamber to evaluate
responsiveness and drift back to lower levels. This work was conducted in the
Cohen Atmospheric Chemistry lab at University of California, Berkeley.
2. The CamPerS then were co-located in Barcelona with ratified government
monitors and with more expensive, research-grade instruments that would
represent a likely choice for conducting limited personal or stationary monitoring
for research studies. In our first deployment, we placed the CamPerS near the
inlet area for the ratified government monitors (Teledyne Monitor Labs 9841 NO2
and NO monitor). The Teledyne Monitor Labs 9841 employ a low pressure
chemiluminescent reaction between ozone and NO to measure the oxides of
nitrogen with typical sensitivity down to ppb levels. Teledyne Monitor Labs
9830B Carbon monoxide analyzer relies on a combination of non-dispersive
7
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
161718
infrared (NDIR) and gas filter correlation (GFC) techniques. For the research
grade instruments, we deployed 2B Technologies Model 410/401 to measure NO2
and NO, and for CO we used the TSI Q-trak model 7565. All of the research-
grade monitors rely on electro-chemical sensors. Due to space restrictions and the
need for direct power supply, research-grade instruments could not be placed near
the inlet area of the government monitors and were instead placed within 3m of
the government monitors and the CamPerS. See Table 1 for a list of the reference
instruments used in the comparison.
3. We deployed the CamPerS in the TAPAS II study alongside several research-
grade instruments. Participants of the study carried the devices with them for an
approximate time of 10 hours, in scripted and free-living exposure conditions.
4. Finally, in the PHENOTYPE study, participants carried the CamPerS for 210
minutes, which included quasi-scripted walks through green spaces, blue spaces,
and urban settings (such as background, residential areas, in transit in
automobiles, and indoors). Participants also had some free-living time wearing the
CamPerS.
Once collected, we converted the millivolts signals from the electro-chemical sensors to
concentrations.The mixing ratio conversion for sensor nodes used the following method:
ppb = (sensor signal (nA) * manufacturer sensitivity (ppb/nA))*calibration factor). The
calibration factor was generated from the recalibration experiments described in the
8
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
192021
second section of the supplementary Appendix II. We then made made corrections for
humidity and temperature as described in the supplementary Appendix I. After applying
the calibration equations and ensuring close matching of time stamps between
instruments, we conducted a series of correlational and agreement analyses. See
Appendix I for details of the various calibrations and corrections applied to the data from
the CamPerS.
We first calculated descriptive statistics to compare each pollutant’s CamPerS
measurements to the government and research-grade monitors. We then examined the
correlations between the CamPers and either the government or research-grade monitors
at 30 minute and 1 hour intervals. In the TAPAS-II study, we had only research-grade
monitors, so we were able to use 1, 5, 10, 30 minute and 1 hour averaging times because
the monitors logged by the minute, rather than the longer averaging times from the
government monitors.
We then combined the CamPerS data from the two studies into microenvironmental
groupings of indoor, free living (indoor and outdoor), urban background, residential low
exposure, green space, blue space, and in-vehicle. We used the urban background as our
comparison group. We ran a generalized linear model (GLM) with 1-minute intervals of
pollution level as the dependent variable and 1-minute indicators of time spent in each
microenvironment as independent predictors. With this modeling framework the
regression coefficients essentially show whether the sensors can differentiate between
different important microenvironments by their sign and magnitude. Sensitivity analyses
9
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
222324
were undertaken using ordinary least squares regression instead of the GLM, but the
results remained largely unchanged and consequently are not reported. In total we
deployed six CamPers repeatedly on a total of 54 participants.
Results
Tables 2A-C show the descriptive statistics for each of the six CamPerS, the government
and the research-grade monitors. Pollution levels estimated by the CamPerS for NO
appear generally to have a positive bias, with average ratios of the CamPerS to
government monitors ranging from 2.32–3.71. The research-grade 2BTech model
410/401 also tended to be biased positively when compared with the Monitor Labs (ML)
model 9841 used at the government site. With the exception of CamPerS C08, most of
the CamPerS monitors have fairly consistent median levels compared to each other, but
the minimum and maximum values show a wide range of variation among different
sensors.
As shown in Figure 2B, for NO2 there was substantial instrument failure due in part to a
prolonged rain event during the deployment. Mean levels of the CamPerS had a positive
bias compared to the government ML 9841, but at the higher concentrations they appear
to be biased on the low side. The research-grade 2BTech 410/401 was generally biased
low compared to the ML 9841 and the CamPerS, although this instrument also failed for
a large portion of the time deployed.
10
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
252627
For CO in Figure 2C, the CamPerS appear to be biased positively on average, but these
biases were quite small compared to the other two pollutants measured. For the maxima,
we see a wide range of values between the CamPerS, and there is some evidence of
positive bias. On average and at the median, however, the CamPerS have readings that
are close to those of the ML model 9830B.
Figure 2 Illustrates the time series of the six CamPerS plotted along with the government
monitors for NO, NO2 and CO. Consistent with the descriptive results, the CamPerS
appear to have positive bias in several instances, but they do detect overall patterns that
are fairly consistent with the government monitors for NO. For NO2 the CamPerS
followed a similar pattern, but there are differences in the levels. Moreover there were
many missing values, which impeded visual inspection. For CO the monitors tend to
follow similar patterns as the government monitors, but did not agree as well as with NO.
Note that the flat lines shown on the government monitor for CO likely correspond to the
lower limit of detection.
As shown in Tables 3A-C, for NO we found moderate to high correlations between the
instruments for averaging times of 30 or 60 minute epochs (rs ~ 0.36 to 0.78).
Correlations with NO2 were low to moderate, ranging from rs ~ 0.04 to 0.67, although
these comparisons were based on far fewer data points. CO correlations were moderate to
high ranging from rs ~ 0.58 to 0.80. In general the correlations were similar for 30 and 60
minute avaraging times, although some variabiation was present. For CO we were unable
to compare with research-grade instruments in this deployment due to equipment failure.
11
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
282930
As shown in the Supplementary Appendix II, we observed moderate to very high
correlations among CamPerS for NO, indicating good replicable measurements (rs ~
0.50-0.90), with many correlations between CamPerS above 0.70. For NO2 the inter-
CamPerS correlations were moderately high but there was limited data available.
Correlations for the CO among monitors were generally moderate to very high (rs ~ 0.59-
0.94), with many correlations above 0.80.
Given the equipment failures on the first field deployment, we redeployed the research-
grade monitors with the CamPerS in February of 2014 during the TAPAS II study. For
the most part, the CamPerS had correlations that were slightly higher for NO with the
research-grade monitors (rs ~ 0.62-0.88) than with the government monitors (see Table
4). Figure 3 shows the time-series plot with 1-minute averaging times from the research-
grade monitors and the CamPerS. Overall the CamPerS follow a similar temporal pattern
to the research-grade instruments.
There were marked improvements in the correlations of NO2 between the CamPerS and
the research-grade 2BTech monitors compared to the correlations with government
monitors (see Table 4). Compared to the correlations with government monitors, we
observed moderate to high correlations between the research-grade and 2BTech monitors
(rs ~ 0.40-0.87). We were also able to examine NOx correlations by summing NO and
NO2 on the CamPerS, and these were also moderate to high (e.g., rs ~ 0.55-0.87 for 10
12
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
313233
minute averages). Temporal patterns for both NO2 and NOx were similar to the research-
grade monitors.
For CO, the correlations were lower than those observed with the government monitors,
in the moderate range of rs ~ 0.23-0.52. Temporal patterns were divergent in some areas
with the CamPerS both over and under-predicting in some periods.
For our next validation, participants carried the monitors in a custom-designed felt case in
various urban microenvironments to determine whether the monitors could detect
meaningful differences between these environments. The custom cases provided some
weather protection for the sensor, but avoided off-gasing from the felt material. We
selected an urban background site as our reference category. Table 5 shows the results for
a generalized linear model, where we controlled for month, time of day, and included a
random effect for each individual for NO and CO. We did not pursue the analysis with
NO2 due to the poor correlations with ratified government monitors, the positive bias, and
the lower volume of recorded data.
Here we see the monitors perform well, detecting significant differences between the
microenvironments. Signs for the coefficients aligned with expectations. Compared to the
urban background, time spent outdoors in green space and blue space was associated with
significantly lower concentrations of CO and NO. We also observed lower concentrations
for time spent in low traffic environments, indoors in a laboratory setting, and in free-
living conditions, which likely included some indoor places. In contrast, time spent in-
13
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
343536
vehicle while traveling to the various sites to conduct the exposure experiments and time
spent in high traffic environments were both associated with higher levels of pollutants.
As noted above, we conducted sensitivity analyses with a linear regression model and
found the results were nearly the same (not shown).
Discussion
After validating the CamPerS in several settings, we found that the CamPerS had good
performance for NO in all validation evaluations. Specifically, they had moderate to high
correlations with ratified government monitors, high correlations with research-grade
instruments, and in linear mixed models they differentiated various micro environments
well. Coefficients all had the expected signs; for example, green space had lower
concentrations compared to the urban background. We did, however, find evidence of
positive bias for the CamPerS compared to the government monitors.
The CamPerS monitors also performed well for CO in our validation. Correlations with
ratified government monitors and research-grade monitors were moderate to high, and we
observed little bias when compared to the government monitors. There was substantial
rainfall during the first deployment, and this reduced the number of valid measurements
for both the CamPerS and the research-grade instruments. In this instance the research-
grade monitors completely failed. In the scripted exposure experiments, the CO
performed similarly to NO sensors, showing good capacity to detect microenvironmental
14
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
373839
differences in exposure that were in line with expectations. The pseudo R-square for the
CO models (0.28), however, was not as large as with the NO model (0.47).
The CamPerS did not perform well for NO2. Correlations with the government monitors
were weak. Interestingly the correlations with research grade instruments were better, in
the moderate to high range. The poor performance compared to government monitors
likely occurred for two reasons. First, the Alphasense Ltd. NO2 sensor is not entirely
specific to NO2. There is a known ozone (O3) interference so that the sensor detects the
combined sum of both O3 and NO2 (see Mead et al. 2013).9 Second, chemiluminescent
monitors used at the government site for measuring oxides of nitrogen (NOx) are often
non-specific in determining NO2 levels. The instruments respond to peroxyacetyl nitrate
(PAN) and several other nitrate species. Readings from the government monitors
probably approximate NOx from the total gas phase rather than the simple sum of NO
and NO2 .12 Whether the poor correlations between the government monitors and the
CamPerS resulted from the non-specificity of government monitors to NO2 or the cross-
interference with O3 in the CamPerS, or both, is difficult to determine with the data
acquired. The relatively high correlations with the 2BTech research-grade monitors and
the CamPerS suggests that at least part of the problem was the non-specificity of the
government monitors that likely measure NOy (i.e., oxides of nitrogen from the total gas
phase) rather than just NO2. In this instance, however, the government monitors would
have likely had a positive bias, yet we still continued to see positive bias with the
CamPerS, so it is unlikely that the problems with the government monitors accounted for
all the low correlations. Earlier analyses conducted in Cambridge, UK, using the same
15
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
404142
CamPerS with co-located government monitors, tended to show higher correlations
between NO2 readings from the government monitors and the CamPerS.9 The comparison
with the government monitor in Cambridge involved NO2 that was O3 corrected. We did
not attempt the O3 correction here because so much of the data was lost during the rain
event and because our scripted experiments involved indoor and in-vehicle time, which
were unlikely to have much O3 present. Consequently it is possible that the poor
correlations between the CamPerS and the government monitors occurred due to O3
interference, problems with the government monitors, or some combination of both.
The sensor manufacturer, Alphasense Ltd, has now developed a new NO2 sensor that has
a scrubber for O3 to deal with the problem of cross-interference (personal communication
with John Saffel, Director of Science, Alphasense Ltd), so that subsequent monitors will
likely have more accurate measures of NO2.
Conclusion
NO and to a lesser extent CO measurements had high correlations among CamPerS,
indicating good replicable measurements (r ~ 0.70-0.99). CamPerS had moderate to high
correlations with government monitors and research-grade instruments for NO,
but had positive bias. For CO, CamPerS had moderate to high correlations with
government monitors and moderate correlations with research-grade instruments, and the
bias was much less compared to NO. The monitors were able to detect high and low air
pollution levels where they would be expected (e.g., high levels on or near busy
16
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
434445
roadways, and lower levels in background residential areas and parks). In general,
primary gases such as NO and CO were measured more accurately than secondary NO2,
which may have had some cross-interference with O3.
The CamPerS demonstrated variable capacity to measure different pollutants when
compared to more expensive monitors, but they did detect meaningful variations in
contrasting microenvironments. Further validations are needed to understand the
conditions under which the sensors provide accurate, reliable information before scaling
up to larger deployments to support epidemiological and citizen science investigations.
Acknowledgements
Ajuntament/Generalitat) who allowed the collocation of CamPerS and provided their data
for us. Funding came from the National Institute of Environmental Health Sciences,
National Cancer Institute, and the HEALS project of the European Commission.
Tables
17
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422423424425426427428429
464748
Table 1. Summary of reference instruments used for comparison with the CamPerS
NO (ppb)Sample Epochs NAs Min. 1st Qu. Median Mean 3rd Qu. Max. Average Ratio
(CamPerS/ML)CamPerS_01 n= 574 264 1.19 2.56 4.10 5.63 7.01 39.78 2.49CamPerS_02 n= 574 232 1.43 3.00 4.37 6.34 7.72 25.92 3.26CamPerS_03 n= 574 388 1.25 3.02 4.50 7.22 8.14 66.04 2.34CamPers_06 n= 574 382 1.02 5.33 6.74 8.14 9.32 39.14 3.60CamPers_08 n= 574 329 1.41 3.92 6.22 9.33 13.76 50.47 3.71CamPers_09 n= 574 330 0.98 3.18 5.03 6.10 7.75 19.66 3.51CamPers_10 n= 574 241 0.91 2.40 3.50 4.76 5.60 25.16 2.32CamPers_12 n= 574 244 0.90 3.23 4.76 6.91 8.19 47.64 3.11CamPers_13 n= 574 276 1.20 3.16 4.70 6.44 7.13 55.44 3.49CamPers_14 n= 574 187 1.19 2.52 3.34 5.11 5.68 37.73 2.52ML (model 9841) n= 574 4 0.76 0.78 1.60 3.40 3.19 47.112BTech (model 410/401) n= 574 161 -2.15 1.60 4.85 6.58 8.95 79.80 2.39
Tables 2A. Comparison of CamPerS to Government and Research-grade monitors showing descriptive statistics - Nitric Oxide Descriptive Comparsions
Note: Sample epochs equals the total number of potential samples during the deployment period, while NA represents the total number of missing observations during the deployment.
NO2 (ppb)Sample Epochs NA's Min. 1st Qu. Median Mean 3rd Qu. Max. Average Ratio
(CamPers/ML)
18
Sensor Nitrogen Dioxide Nitric Oxide Carbon Monoxide
GovernmentMonitors
Monitor Labs 9841 Monitor Labs 9841 Monitor Labs 9830B
Research-grade Instruments
2BTechnologiesModel 410/401
2B Technologies Model 410
TSI Q-trak model 7565
431432433434435436
437438439440441442443444445446447
495051
CamPerS_01 n= 574 359 4.03 9.74 16.02 15.94 22.24 31.12 2.06CamPerS_02 n= 574 498 6.11 10.39 12.94 16.30 21.74 38.67 1.55CamPerS_03 n= 574 - - - - - - - -CamPerS_06 n= 574 - - - - - - - -CamPerS_08 n= 574 490 4.51 11.39 15.16 16.42 21.96 33.37 3.54CamPerS_09 n= 574 - - - - - - - -CamPerS_10 n= 574 - - - - - - - -CamPerS_12 n= 574 - - - - - - - -CamPerS_13 n= 574 - - - - - - - -CamPerS_14 n= 574 369 4.97 17.00 26.20 24.41 31.78 66.14 2.89ML (model 9841) n= 574 4 0.51 6.26 11.89 14.97 20.02 53.18 1.002BTech (model 410/401) n= 574 161 -0.76 1.93 3.83 5.03 6.79 24.80 0.32
Tables 2B. Comparison of CamPerS to Government and Research-grade monitors showing descriptive statistics - Nitrogen Dioxide Descriptive Comparsions
CO (ppb)Sample Epochs NA's Min. 1st Qu. Median Mean 3rd Qu. Max. Average Ratio
(CamPers/ML)CamPerS_01 n= 574 278 0.01 0.08 0.18 0.24 0.33 1.30 1.09CamPerS_02 n= 574 231 0.01 0.12 0.21 0.28 0.37 1.36 1.29CamPerS_03 n= 574 387 0.01 0.11 0.23 0.30 0.40 1.23 1.24CamPerS_06 n= 574 381 0.01 0.11 0.20 0.24 0.33 0.81 1.06CamPerS_08 n= 574 328 0.01 0.12 0.24 0.28 0.41 1.08 1.19CamPerS_09 n= 574 327 0.01 0.14 0.22 0.29 0.38 1.45 1.38CamPerS_10 n= 574 240 0.01 0.18 0.29 0.36 0.46 2.12 1.64CamPerS_12 n= 574 245 0.01 0.11 0.20 0.29 0.40 1.48 1.34CamPerS_13 n= 574 276 0.02 0.08 0.16 0.23 0.30 1.12 1.06CamPerS_14 n= 574 188 0.01 0.11 0.20 0.26 0.34 1.16 1.20ML (model 9830B) n= 574 6 0.16 0.17 0.17 0.20 0.25 0.59 -
Tables 2C. Comparison of CamPerS to Government and Research-grade monitors showing descriptive statistics - Carbon Monoxide Descriptive Comparsions
30 minutes 60 minutes
NO (ppb)ML
(model 9841)2BTech
(model 410/401)ML
(model 9841)2BTech
(model 410/401)
19
448449450451452453454455456
457458459460461
525354
CamPerS_01 0.65 0.62 0.65 0.61CamPerS_02 0.57 0.47 0.55 0.48CamPerS_03 0.78 0.73 0.78 0.76CamPerS_06 0.57 0.55 0.59 0.66CamPerS_08 0.51 0.45 0.48 0.45CamPerS_09 0.47 0.41 0.44 0.38CamPerS_10 0.53 0.39 0.52 0.43CamPerS_12 0.43 0.38 0.44 0.36CamPerS_13 0.53 0.69 0.55 0.65CamPerS_14 0.63 0.52 0.64 0.492BTech(model 410/401) 0.83 0.85
Tables 3A-C. Summary of Spearman’s correlations between the CamPerS and government or research-grade monitors with 30 min and 1 hour averaging
3A. Nitric oxide correlations between CamPerS and either research-grade or government monitors
30 minutes 60 minutes
NO2 (ppb)ML
(model 9841)2BTech
(model 410/401)ML
(model 9841)2BTech
(model 410/401)CamPerS_01 0.12 0.56 0.11 0.56CamPerS_02 0.53 0.47 0.55 0.51CamPerS_08 0.19 0.16 0.16 0.04CamPerS_14 0.52 0.67 0.52 0.702BTech (model 410/401) 0.94 0.92
3B. Nitrogen dioxide correlations between CamPerS and either research-grade or government monitors
30 minutes 60 minutes
CO (ppm)ML
(model 9830B)Q-trak
(model 7565)ML
(model 9830B)Q-trak
(model 7565)CamPerS_01 0.67 NA 0.70 NA
20
462463464465466467468469
470471472473474475476477478479480481
555657
CamPerS_02 0.59 NA 0.58 NACamPerS_03 0.80 NA 0.82 NACamPerS_06 0.68 NA 0.69 NACamPerS_08 0.68 NA 0.66 NACamPerS_09 0.64 NA 0.66 NACamPerS_10 0.59 NA 0.58 NACamPerS_12 0.73 NA 0.74 NACamPerS_13 0.74 NA 0.76 NACamPerS_14 0.68 NA 0.68 NA
3C. Carbon monoxide correlations between CamPerS and either research-grade or government monitors
CamPerSIndividual
Monitors for Gaseous Species
rs Between Research
and CamPerS(1 min avg)
rs Between Research
and CamPerS(10 min avg)
rs Between Research
and CamPerS(30 min avg)
rs Between Research
and CamPerS(60 min avg)
NO - - - -CamPerS_C01 0.67 0.68 0.65 0.65CamPerS_C02 0.81 0.83 0.82 0.70CamPerS_C03 0.75 0.78 0.78 0.86CamPerS_C08 0.76 0.79 0.79 0.88CamPerS_C09 0.81 0.85 0.87 0.93CamPerS_C10 0.81 0.85 0.82 0.81CamPerS_C14 0.80 0.83 0.85 0.93
NO2 - - - -CamPerS_C01 0.40 0.47 0.43 0.45CamPerS_C02 0.78 0.87 0.82 0.60CamPerS_C03 0.41 0.50 0.46 0.45CamPerS_C08 0.55 0.67 0.72 0.83CamPerS_C09 0.61 0.73 0.69 0.79CamPerS_C10 0.66 0.75 0.72 0.74CamPerS_C14 0.62 0.72 0.73 0.83
NOxCamPerS_C01 0.61 0.61 0.56 0.56CamPerS_C02 0.75 0.77 0.68 0.56CamPerS_C03 0.68 0.75 0.77 0.81CamPerS_C08 0.71 0.82 0.85 0.94CamPerS_C09 0.77 0.86 0.84 0.91CamPerS_C10 0.80 0.85 0.84 0.83CamPerS_C14 0.79 0.84 0.87 0.95
21
482483484485486
585960
COCamPerS_C01 0.45 0.53 0.52 0.48CamPerS_C02 0.45 0.50 0.48 0.44CamPerS_C03 0.23 0.29 0.25 0.18CamPerS_C08 0.32 0.36 0.33 0.28CamPerS_C09 0.36 0.43 0.42 0.35CamPerS_C10 0.39 0.42 0.41 0.35CamPerS_C14 0.41 0.48 0.47 0.38
Table 4. Summary of Spearman’s correlations between the CamPerS and research-grade monitors with 1 min, 10 min, 30 min and 1 hour averaging for the TAPAS-II Redeployment
NO (ppb) CO (ppm)
Coefficient (95% CI) Coefficient (95% CI)
Intercept (Urban Background) 55.6 (46.1 , 65.2) 1.46 (1.37 , 1.55)
Blue Space -32.3 (-35.5 , -29.1) -0.68 (-0.73 , -0.63)
Green Space -32.8 (-36.0 , -29.6) -0.71 (-0.76 , -0.66)
Free Living 3.6 (1.0 , 6.3) -0.17 (-0.21 , -0.13)
In-vehicle 89.4 (85.1 , 93.6) 1.27 (1.2 , 1.33)
Indoors in Lab Setting -14 (-17.2 , -10.7) -0.25 (-0.30 , -0.20)
Low Traffic -34.6 (-38.2 , -30.9) -0.46 (-0.52 , -0.40)
High Traffic 284 (280.3 , 287.8) 2.69 (2.64 , 2.75)Pseudo R-square 0.47 0.28
Table 5. Results of the generalized linear models with measured pollution as the dependent variable and time spent in microenvironments as the predictors
Note: Both models control for month and time of day. Pseudo R-square was caculated based on methods contained in Gurka (2006).
22
487488489490491
492493494495496497498499500501502503504505
616263
Figures
Figure 1. Cambridge Personal Sensor (CamPerS) with essential components shown (approximate weight 450 grams with batteries and 330 grams without batteries)
23
506507
508509510511512513
646566
24
514
515
676869
Figure 2. Time series plots of each pollutant with CamPerS versus government monitors for NO, NO2 and CO with 30 minute averaging times
25
516517518519520
707172
26
521
522
737475
Figure 3. Time-series plots of each pollutant with CamPerS versus research-grade monitors (NO, NO2, NOx, CO)
27
523
524525526527
767778
References
(1) Jerrett, M.; Arain, A.; Kanaroglou, P.; Beckerman, B.; Potoglou, D.; Sahsuvaroglu, T.; Morrison, J.; Giovis, C. A Review and Evaluation of Intraurban Air Pollution Exposure Models. J. Expo. Anal. Environ. Epidemiol. 2005, 15 (2), 185–204.
(2) Zeger, S. L.; Thomas, D.; Dominici, F.; Samet, J. M.; Schwartz, J.; Dockery, D.; Cohen, A. Exposure Measurement Error in Time-Series Studies of Air Pollution: Concepts and Consequences. Environ. Health Perspect. 2000, 108 (5), 419–426.
(3) Suh, H. H.; Zanobetti, A. Exposure Error Masks the Relationship Between Traffic-Related Air Pollution and Heart Rate Variability: J. Occup. Environ. Med. 2010, 52 (7), 685–692.
(4) National Academy of Science Committee on Human and Environmental Exposures. Exposure Science in the 21st Century: A Vision and a Strategy (2012). Washington DC: National Research Council.
(5) Lahoz, W. CITIZENS’ OBSERVATORY - Where Does Our Air Quality Come From? Meteorol. Int. August, 126–128.
(6) Turner M.C., Nieuwenhuijsen M., Anderson K., Balshaw D., Cui Y., Dunton, G., Hoppin J.A., Koutrakis, P., Jerrett M. 2016. Assessing the Exposome with External Measures: Commentary on the State of the Science and Research Recommendations. Annu. Rev. Public Health. In press.
(7) Liu, H.-Y.; Kobernus, M.; Broday, D.; Bartonova, A. A Conceptual Approach to a Citizens’ Observatory – Supporting Community-Based Environmental Governance. Environ. Health 2014, 13, 107.
(8) Nieuwenhuijsen, M. J.; Donaire-Gonzalez, D.; Rivas, I.; de Castro, M.; Cirach, M.; Hoek, G.; Seto, E.; Jerrett, M.; Sunyer, J. Variability in and Agreement between Modeled and Personal Continuously Measured Black Carbon Levels Using Novel Smartphone and Sensor Technologies. Environ. Sci. Technol. 2015, 49 (5), 2977–2982.
(9) Mead, M. I.; Popoola, O. A. M.; Stewart, G. B.; Landshoff, P.; Calleja, M.; Hayes, M.; Baldovi, J. J.; McLeod, M. W.; Hodgson, T. F.; Dicks, J.; et al. The Use of Electrochemical Sensors for Monitoring Urban Air Quality in Low-Cost, High-Density Networks. Atmos. Environ. 2013, 70, 186–203.
(10) Donaire-Gonzalez, D.; de Nazelle, A.; Seto, E.; Mendez, M.; Nieuwenhuijsen, M. J.; Jerrett, M. Comparison of Physical Activity Measures Using Mobile Phone-Based CalFit and Actigraph. J. Med. Internet Res. 2013, 15 (6), e111.
28
528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572
798081
(11) de Nazelle, A.; Seto, E.; Donaire-Gonzalez, D.; Mendez, M.; Matamala, J.; Nieuwenhuijsen, M. J.; Jerrett, M. Improving Estimates of Air Pollution Exposure through Ubiquitous Sensing Technologies. Environ. Pollut. 2013, 176, 92–99.
(12) Winer, A. M.; Peters, J. W.; Smith, J. P.; Pitts, J. N. Response of Commercial Chemiluminescent Nitric Oxide-Nitrogen Dioxide Analyzers to Other Nitrogen-Containing Compounds. Environ. Sci. Technol. 1974, 8 (13), 1118–1121.
29
573574575576577578579580581582583584
828384
