This is the PhD defense presentation of Luc Dekoninck. It presents an innovative methodology to extend land-use regression into personal exposure modeling. In the presented cases, noise exposure is used as a proxy for exposure to traffic. The increased resolution of the models enables the disentanglement of local and background contributions in the personal exposure to Black Carbon.
Luc.
Spatiotemporal modeling of personal exposure to particulate matter
1. Spatiotemporal modelling of
personal exposure to traffic related
particulate matter
using noise as a proxy
Luc Dekoninck
1Acoustics Group, Department of Information Technology
Ghent University, Belgium
3. Context: Traffic related Quality of Life
3
Traffic Liveability
• Person centered approach
• Include trip behaviour
Traffic livability
HealthAccessibility of
functions
Living
environment
Social
cohesion
Shops, schools,
employment,
recreation,
…
Road safety
External effects:
traffic noise
air quality
Disturbance
traffic noise
odour
Public areas
visual disturbance
matching soundscape
Subjective road
safety
Barrier effect
Facilities for
pedestrians and
cyclists
Aspects
Travel time / cost
“Weighed travel
time”, …
Accident risk,
DALY / QALY, …
Noise exposure,
Odour exposure
Density of traffic
Traffic flow
Crossing time
…
Indicators
Traffic
impacts
Goal: predict the response to a QoL question
based on geographical information only
Geographic information for modeling is a discipline/technique:
Land-Use regression modeling or LUR
(verkeersleefbaarheid)
4. People evaluate traffic through noise
4
Perceived QoL (subjectieve verkeersleefbaarheid)
= function(noise@dwelling, noise@nearby roads, …)
Bustle = local traffic
perceived
through
auditory
system
Noise
Direct path
5. Land-
5
Noise as a proxy for QoL… and maybe more
Noise Traffic Air pollution
Proxy
Perceived QoL
= function(noise, air pollution, safety, walkability…)
PhD Goal:
Air pollution exposure (Particulate matter)
= function(noise@dwelling, noise@along roads, …)
?
6. 6
Vehicle noise emission and vehicle BC/UFP
emission are physically linked
Cruising (low speed)
Vehicle emissions
PM
(BC/UFP)
Cruising (high speed)
Accelerating (low speed)
Decelerating/Idle
=
-
+
+++
Engine
noise
=
-
+++
+
Rolling
noise
=
=/-
+
+++
The physical link is the engine regime
Total
noise
=
=/-
+++
+++
8. Bicyclists’ exposure to Black Carbon:
One year of data in Ghent
8
2011
200 + trips
2500 km
> 75 km distinct roads
Random routes
All weather
except rain
µAethalometer
Noise level meter
(1/3 octave bands)
GPS
BC (µg/m3)
9. Noise includes traffic dynamics
NoiseLevel(dBA)
Freq (Hz)
High speed, no acceleration
Low speed, acceleration
OLF
Traffic count
+ engine regime
+ distance to source
OHF
Speed related
HFmLF = OHF – OLF
speed indicator
relative to OLF
HFmLF
Engine Tyre
13. Bicycle BC exposure:
fitting the trips is succesfull
13
Dekoninck L., Botteldooren D. & Int Panis L. 2013.
An instantaneous spatiotemporal model to predict a bicyclist's Black Carbon exposure
based on mobile noise measurements. Atmospheric Environment, 79, 623-631.
Each dot is a single trip
15. Particulate matter in general
15
EU legislation
Health
All soot/black/combustion
products
16. Personal exposure to BC
16
In-traffic BC exposure:
20% to 60% of the daily BC exposure
(Dons et al., 2011/2012)
This dataset wil be used
as external validation
Personal
Exposure
Includes all activities
Sensitive to personal behaviour
19. Activity and dose…
19
Activity
Inhalation rate
Lung volume
Particle size
Deposition
internal dose
specific organs
Chemical loading
Fitness
The models do not include any dose correction:
external exposure only
20. Equipment:
Cheap (fixed) noise measurement system
Extended with µAethalomter and GPS
= Mobile version of the hardware build
in IWT funded IDEA project
Can the same approach be used for UFP?
Extreme traffic situation in Bangalore, India…
20
21. Background adjusted approach is
valid for BC and UFP
21
High background hampers the modelUFPloc
Honking
Dekoninck, Luc, Dick Botteldooren, Luc Int Panis, Steve Hankey, Grishma Jain, Karthik S, Julian Marshall.
“Applicability of a noise-based model to estimate in-traffic exposure to black carbon
and particle number concentrations in different cultures” Environment International, Volume 74, January 2015, Pages 89–98.
23. Yearly averaged exposure for policy and health
23
Trips (+ traffic assessment by noise)
Predictions for all
meteorological conditions
for all trips
Aggregate
Yearly averaged
local exposure
Apply
instantaneous model
24. How efficient is the mobile noise sampling?
24
Four passages identify
the local traffic situation
and result in
an annual BC estimate
Within 500 ng/m3
25. Application:
City wide traffic mapping through noise
supplies detailed traffic and traffic dynamics
and results in city wide BC exposure
25
Important application
for all mobility related disciplines !!
27. In-vehicle exposure to Black Carbon
One year of data in Flanders
27
2013
340 + trips
> 220 hours
9 volunteers
Random routes
All weather
µAethalometer
Mobile node
Noise map as LUR traffic layer ?
29. In-vehicle exposure model
noise map based, six parameters
29
Hour of Day Noise map Wind speed Temperature Background Street Canyon
30. In-vehicle exposure: external validation
(validation data of 2010-2011, measurements in 2013)
30
Emission limits (euro5, sept 2009):
35-40% emission reduction feasible
Fleet emission correction of 40%
Good correlation
Bad absolute levels
Missing over 50%
of the exposure
Near road concentration during rush
of government show
40% emission reductions
34. 34
Activity specific indicator:
a function of personal information,
activity specific features
and environmental attributes
Temporal resolution of the ASMs:
function of the variability of the activity
35. In-vehicle BC exposure:
Step 1: Participatory sensing + Land-use
35
Activities: car trips (BC + GPS)
225 hours of data
9 individuals
10 second resolution
External (Land-use) data:
meteo, traffic data,
background concentrations,
noise map, traffic dynamics,
build-up area,…
Dataset to explore
and build a model
ASF is
unknown
Proposed ASF
36. In-vehicle exposure:
Step 2: External validation
36
Person factory:
external car trips
(BC + GPS)
BC in 5 minutes resolution
ASF to
test / validate
Prediction
by ASF
External validation
dataset
Choose evaluation level
Identical
attribution
Valid ??
38. Does the background adjusted approach
work for indoor exposure?
38
Bkg
Local
BC background time series PM10 regional map
Yearly BC 5.5 - 7.0 % of PM10
regional noise map
39. Bicycle model downgrade:
shift from LOLF to LAeq, traffic
39
Gam model (strong reduction in traffic data and no traffic dynamics included)
𝑩𝑪 𝑩𝑨,𝒐𝒖𝒕𝒅𝒐𝒐𝒓 = 𝑩𝑪 𝒃𝒌𝒈 + 𝑩𝑪𝒍𝒐𝒄,𝒕𝒓𝒂𝒇
𝑩𝑪 𝑩𝑨,𝒊𝒏𝒅𝒐𝒐𝒓 = 𝑩𝑪 𝑩𝑨,𝒐𝒖𝒕𝒅𝒐𝒐𝒓 𝑰/𝑶 𝒓𝒂𝒕𝒊𝒐(temperature)
40. Relative fit across the covariates
(predicted BC / BC measurement)
40
Fixed I/O correction from literature (0.78) Daily temperature
Retrofit resolves the trend
in the background exposure
Retrofit
In line with literature
After temperature adjustement
RelativefitRelativefit
41. Indoor exposure model:
external validation
41
Restrictions in the input data:
• Large reduction in the resolution of traffic data
• No traffic dynamics
• No specific diurnal pattern for different dwellings
• General (seasonal) I/O correction, no dwelling specific information
Each dot is an individual activity
43. Background adjusted model:
BC exposure at a dwelling façade
43
Busy street
Busses
Street Canyon
Six weeks
Easter holiday
Large traffic variation
Large meteo variation
Large bkg variation
44. How to evaluate the quality of the
background adjustment
44
Raw data
at Tolhuis
Correction
Baudulo Park
Correction fails…
Closest background
Baudulo Park
45. Back
45
Background adjusted model:
Also valid at a dwelling façade ?
Raw data
BCbkg OLF Wind Temp
Baudulo
Correction
with Baudulo
Baudulo
Correction
with Antw-LinkOever
AntwLink
Correction
with Houtem-Veurne
Houtem
Link-Oever
adjustement
better but still…
Houtem
adjustement
Model with
potential
46. Back
46
Single facade pilot model
Background adjusted
Local city contribution is missing
Let’s add the local contribution
Blue =
Houtem + local city adjustment
48. The (instantaneous) daily exposure model
48
+ walk (+ rail + light rail)
few activities
+ bus
+ indoor@other destinations
ASM for bicyclists:
BCtot = BCbkg + gam(Lday, wind speed, street canyon index)
ASM for in-vehicle:
BCtot = gam(Lday, wind speed, Temp, HourOfDay, BCbkg, stcan)
ASM for indoor@home:
BCtot = (BCbkg + gam(Lday, wind speed, street canyon index)). I/O(temp24h)
49. Daily exposure: external validation
49
293 person-days, summer and winter
No information of external campaign was used in the models
50. Indoor Black Carbon sources?
50
Non-smokers
No evidence
of indoor BC sources
Other activities
→ Other I/O ratio ?
→ Other BC sources ?
→ Outdoor ?
Relative fit by Purpose
@ Home
Other
51. Evaluation by micro-environment
51
Bike & Walk
Underestimated
Lack of local traffic and traffic dynamics
Potential solution
City-wide mapping
This is an upgrade
53. The microscopic and micro-environment
specific non-linear spatiotemporal
land- use regression model (µLUR)
53
2. Measurement campaign designed to capture as much of
the variability over all driving forces
1. Low aggregation level of the measurements
5. Instantaneous non-linear (gam) models (for ln(BC))
Features:
3. Spatial and temporal attribution of all driving forces
4. Activity and/or micro-environment specific models
Traffic Related Air Pollution = traffic + traffic dynamics captured through noise
55. Add dose…
55
Activity
Inhalation rate
Lung volume
Particle size
Deposition
internal dose
specific organs
Chemical loading
Fitness
The models do not include any dose correction:
external exposure only
57. Personal exposure for
health effects
57
Behavior
of individuals in cohort
ASMs
are
known
Personal Exposure
Health outcomes
Biomarkers
Effects ??
58. Personal exposure for
policy support
58
Simulated behavior
of a population
of individuals
ASMs
are
known
Population
distribution of
selected indicators
Evaluate different policy scenario’s
• Modal shift
• Improved bicycle network
• Additional focus on alternative routes
• Reduce traffic demands
• …
59. Conclusions
• Traffic assessment using noise as a proxy is successful
• Instantaneous traffic assessment enables the
disentanglement of the BC/UFP exposure into a local and a
background component
• Validated Activity Specific Models
• Bicyclists; In-vehicle and @home
• External validation of
Personal Daily BC exposure is successful
• Huge potential synergies between
noise, traffic, air pollution
and epidemiologic disciplines possible
• Models applicable on any mobile population for advances in
epidemiology and policy support
59