04/10/20128th AIAI Conference, 27 – 30 September, Halkidiki, Greece
Tutorial COST Action TD1105 EuNetAir
Presentation: Jan Theunis
New approaches in outdoor air quality
monitoring: mobile sensing, participatory
sensing and sensor networks
Research team:
Matteo Reggente
Jan Peters
Martine Van Poppel
Evi Dons
Joris Van den Bossche
04/10/2012 2© 2012, VITO NV
Outline
» Air quality monitoring
» New approaches: focus on exposure and health
» Sensor networks: concept, examples of statistical modelling
» Mobile monitoring: tools and methods
» Participatory monitoring: sesnor array
04/10/2012 3© 2012, VITO NV
Urban air quality0
10
20
30
40
50
60
70
80
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
NO2_hour ly_average
CO_hourly_average
O3_hourly_average
0
10
20
30
40
50
60
70
80
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
NO2_hourly_average
CO_hourly_average
O3_hourly_average
summer
winter
NO2
O3
CO
NO2
O3
CO
04/10/2012 4© 2012, VITO NV
NOxCO
VOC
O3
UFP
BC
?? ?
Exposure: people moving through the
environment
» Dynamic Exposure Assessment
» Need for detailed data with high spatial and temporal resolution
04/10/2012 5© 2012, VITO NV
Air quality monitoring
Air quality is not homogeneous in urban environments!
154000 154100 154200 154300 154400 154500 154600 154700 154800 154900
211000
211100
211200
211300
211400
211500
211600
211700
211800
211900
1 Urban air quality station for Antwerp
- Conventional: Reference methods- Only regulated components- “Correct” but poor spatial coverage
04/10/2012 6© 2012, VITO NV
New approaches: focus on exposure and health
» health-relevance versus regulation
» exposure in different micro-environments
» detailed data – high spatio-temporal resolution
» Sensor networks : low cost sensors
» Mobile monitoring: high quality monitors – low cost sensors
» Participatory monitoring
» Challenges: sensor quality, data quality, mobile data, intelligent data
processing
04/10/2012 7© 2012, VITO NV
Sensor networks: concept
» Heterogeneous networks of low cost sensors and high performance
instrumentation
» Lack of accuracy compensated by amount of data
» Making use of network intelligence (incl. learning capabilities) to
guarantee overall quality
» Combined measurement of different agents, e.g. different air pollutants,
noise, meteo, …
» Making sensors work closely together
» UFP, NOx, CO, noise � “proxies”
» Data aggregation and mining
04/10/2012 8© 2012, VITO NV
Measuring device- 1000 - 2000 €
Basic sensors- Electrochemical- Semiconductor metaloxide- 5 – 80 €
Sensor head- temperature control- calibration curve- correcting for T, RH- 200 - 300 €
Not designed for / littleexperience in ppb range cross-interference, drift,
T and Hum effects
Sensor networks: low cost sensors
04/10/2012 9© 2012, VITO NV
Sensor networks: low cost sensors
• Sensor node with low-cost CO sensor:
• � Huge temperature dependence
Sensor Electronics
Outside temperature and relative humidity
CO sensor temperature
CO sensor
04/10/2012 10© 2012, VITO NV
• Sensor node with low-cost CO sensor
» correcting interference of environmental factors
» Test set-up: sensors collocated with reference CO monitor
» Develop statistical model
Sensor networks: low cost sensors
Two weeks
Two weeks
04/10/2012 11© 2012, VITO NV
Sensor networks: low cost sensors
• Sensor node with low-cost CO sensor» correcting interference of environmental factors: result
04/10/2012 12© 2012, VITO NV
Site 1
Site 2
Site 3
» Estimating UFP (ultrafine particles)» With the help of NO and NO2 measurements
NOx meas.
UFP pred.UFP
NOx meas.
UFP pred.UFP
NOx meas.
UFP pred.UFP
Sensor networks: sensors learning
from each other
Matteo Reggente, Jan Theunis, et al (in
preparation) Prediction of Ultrafine
Particles Number Concentration by
means of NO, NO2 and Gaussian Process
in Urban Environment
04/10/2012 13© 2012, VITO NV
Sensor networks: sensors learning
from each other: locations and measurements
Site the distance from traffic (m)
Weekday traffic volume (vehday–1)
Weekend traffic volume (vehday–1)
Heavy duty fraction on weekday (and weekend)
(%)
Site 3 20-30 37,000 25,000 7% (3%)Site 1 3 5,000 4,000 5% (2%)Site 2 2 4,000 3,000 4% (2%)
AirPointer chemoluminiscence monitor ���� NO, NO2Grimm NanoCheck ���� UFP Concentrations (25-300nm)
04/10/2012 14© 2012, VITO NV
Sensor networks: sensors learning
from each other
04/10/2012 15© 2012, VITO NV
Sensor networks: sensors learning
from each other: Statistical Modelling
IGMM
Training Input:Time
NO, NO2
Evaluation Input:Time
NO, NO2
C1
C2
C3
Cn
...
GP 1
...
GP 2
GP 3
GP n
Training Target (UFP)
UFP1
UFP2
UFP3
UFPn
Merge M
ixtures...___UFP
UFP Measured
Two weeks
Two weeks
Using half hour averages
04/10/2012 16© 2012, VITO NV
Statistical modelling: Evaluation
Time Series and Scatter Plot:
Measured versus estimated UFPconcentrations
04/10/2012 17© 2012, VITO NV
Statistical modelling: Evaluation
Under- prediction
Empirical cumulative distribution and error distribution
04/10/2012 18© 2012, VITO NV
Mobile air quality monitoring
» Objectives :
» Obtain spatially and temporally resolved data on air quality
» Applications :
» Personal exposure monitoring» Berghmans P, Bleux N, Int Panis L, Mishra V, Torfs R, Van Poppel M, 2009. Exposure assessment of a cyclist to
PM10 and ultrafine particles. Science of The Total Environment, Volume 407, Issue 4, 1286-1298
» Hot-spot identification: mapping in urban and industrial environments
to assess impact of local sources
» High resolution mapping in urban environment
» Data acquisition for model calibration
04/10/2012 19© 2012, VITO NV
» Portable instrument
» Micro-aethalometer AE51
» Black carbon
» Electronic diary
Dynamic exposure assessment: personal
exposure monitoring
Dons E, Int Panis L, Van Poppel M, Theunis J, Willems H, Torfs R, Wets G
(2011), Impact of time-activity patterns on personal exposure to black
carbon, Atmospheric Environment, Volume 45, Issue 21, July 2011, p.
3594-3602,
Dons, E., Int Panis, L., Van Poppel, M., Theunis, J., & Wets, G. (2012).
Personal exposure to Black Carbon in transport microenvironments.
Atmospheric Environment 55, 392-398. Average proportion of time spent on activities (lef t) and corresponding proportion of black carbon exposure p er activity (right)
04/10/2012 20© 2012, VITO NV
Aeroflex – Air Quality Bike
Display with ‘traffic lightGUI’
UFP: TSI P-Trak
Netbook and batteries
PM1, PM2,5, PM10: Grimm 1.108GPS: GlobalSat BU-353
UFP probe
PM probe
Bart Elen, Jan Peters, Martine Van Poppel, Nico Bleux,
Jan Theunis, Matteo Reggente, Arnout Standaert
(submitted 2012) The Aeroflex: a bicycle for mobile air
quality measurements, submitted to Sensors
04/10/2012 21© 2012, VITO NV
Aeroflex Data Infrastructure - Overview
Pre-processed data
Step 1: Measurementenhancement
Step 2: Datavalidation
Step 4: Dataaggregation
RAW data
3. Automated data pre-processing chain
CSVJSON
1. Data transmission and storage
4. Data visualizationand analysis
2. Quick data visualization
Step 3: Backgroundcorrection
collect lots of mobile measurements
-> requires automated data processing
04/10/2012 22© 2012, VITO NV
Aeroflex Data Infrastructure – Need for
adaptability» Must be ready to adapt set of measurement devices:
GRIMM agentGRIMM agent
Data logger & transmitter agent
Data logger & transmitter agent GUI agentGUI agent
GPS agentGPS agent µ Aeth agentµ Aeth agent Center 322 noise agentCenter 322 noise agent
TinyTag agentTinyTag agent …
Measurement device agents
P-Trak agentP-Trak agent
OSGi services
In hardware: USB-network
In sofware: Loosely coupled software agents
Implemented as OSGi
bundles
04/10/2012 23© 2012, VITO NV
Mobile monitoring: methodology
» Spatio-temporal data
» L = {time, location, air quality}
» Single run: snap shot - Highly influenced by
traffic discontinuity and short term incidents
» Spatio-temporal series of measurements
» Fixed route
» Repeated measurements + data
aggregation
» Background correction
Spatio-temporal series
Repeated measurements: one morning
Provinci Straat-31-03-09-N1
050000
100000150000200000250000300000350000400000
10:4
710
:47
10:4
710
:47
10:4
710
:48
10:4
810
:48
10:4
810
:49
10:4
910
:49
10:4
9
Time (HH:MM)
Num
bers
cm
-3
UFP number concentration Provinciestraat during a single passage
04/10/2012 24© 2012, VITO NV
Case study
» Two locations: Antwerp (medium-sized city, 480 000 inhabitants, 985
inhabitants km-2) and Mol (provincial town, 34 000 inhabitants, 300
inhabitants km-2)
» Fixed route at both study sites :
» 24 runs in Antwerp, 8 dates in the period between March 16 and April 8, 2009
» 20 runs in Mol, 10 measurement dates between April 7 and April 23, 2010
» Measurement times
Antwerp (5 km) Mol (10 km)
04/10/2012 25© 2012, VITO NV
Methodology : street level aggregation
Street level aggregation of all data
of the Mol route (above) and the
Antwerp route (below) show
significant differences in UFP
concentrations between streets
Jan Peters, Jan Theunis, Martine Van Poppel,
Patrick Berghmans (submitted 2012) Monitoring
PM10 and ultrafine particles in urban
environments using mobile measurements,
submitted to Aerosol and Air Quality Research
04/10/2012 26© 2012, VITO NV
Methodology : street level aggregation
» Exponential decay
» Distance Aeroflex – traffic important» � Restrictions for use of Aeroflex : measurements are
representative in the first place for the pollutant
concentrations that cyclists are exposed to.
43 000 day-1 13 000 day-1
Plantin & Moretuslei Provinciestraat
04/10/2012 27© 2012, VITO NV
Methodology : traffic discontinuity and short-
term incidents» Data aggregation → representativeness
» Gaussian kernel smoothing of spatio-temporal data
» Level out part of the variability that is related to traffic discontinuity
and short-term incidents
» Smooth accumulated data (eg. at traffic light)
04/10/2012 28© 2012, VITO NV
Participatory monitoring
• detailed spatial and temporal scale
dynamic exposure assessment
• corresponding to people’s personal
environment and activities
• collaborative efforts to collect large
representative datasets for mapping
urban environment
• enhance people’s understanding of the
urban environment
• contribute to collaborative decision-
making processes
04/10/2012 29© 2012, VITO NV
Participatory sensing – Portable air quality
monitoring devices
» Opportunistic sensing: people collecting data during their normal daily
routine
3 groups of ‘City Guards’ measuring
air quality in Antwerp during 6 months
BC Mapper
Home station:
- reading out the data
- clock synchronisation
- send data to database
- recharge equipment
Portable air quality monitoring
kit with minimal impact on
volunteers
User friendly !
04/10/2012 30© 2012, VITO NV
» https://sites.google.com/site/urbanbcmeasurements/home
» Challenges:
» GPS corrections and exact locations
» Indoor versus outdoor
» Interferences, e.g. smoking
Participatory sensing – Portable air quality
monitoring devices
04/10/2012 31© 2012, VITO NV
0
0,2
0,4
0,6
0,8
1
1,2
1,4
1,6
0
0,5
1
1,5
2
2,5
3
3,5
3/05/2011 13:12 3/05/2011 13:26 3/05/2011 13:40 3/05/2011 13:55 3/05/2011 14:09 3/05/2011 14:24 3/05/2011 14:38 3/05/2011 14:52 3/05/2011 15:07 3/05/2011 15:21 3/05/2011 15:36
PPM CO Sid 180 CO TGS 2442 Sid 3102 TGS 2201 Gasoline Sid 3202 NO2 MiCS 2710 Sid 3101
Low cost measurement devices for large scale
data collection: EveryAware SensorBox
� poor selectivity
� cross-interference
� drift
� T and humidity effects
Individual low-cost sensors Sensor array (e-nose)
Response of 4 different gas sensors to traffic-related pollution
� Exploit partial selectivity
towards different gas
components using machine
learning tools to achieve
multivariate calibration
04/10/2012 32© 2012, VITO NV
Low cost measurement devices for large scale
data collection: EveryAware SensorBox
10 sensor e-nose- 7 sensors which react on traffic pollution-Ozone, Temperature and Relative humidity for sensor correction
Bart Elen, Jan Theunis, Stefano Ingarra, Andrea Molino, Joris Van den Bossche, Matteo Reggente and Vittorio Loreto (2012) The
EveryAware SensorBox: a tool for community-based air quality monitoring, paper presented at the Workshop Sensing a Changing
World, May 9-11, 2012, Wageningen, The Netherlands. (http://www.geo-
informatie.nl/workshops/scw2/papers/Elen_etal_EveryAware_SensorBox.pdf )
04/10/2012 33© 2012, VITO NV
» Deployment of the Sensor Boxes close to a monitor device
» Use them both stationary and mobile
Low cost measurement devices for large scale
data collection: Multivariate Calibration
04/10/2012 34© 2012, VITO NV
» Deployment of a subset of Sensor Boxes close to portable
and “TRUSTABLE” device in a mobile contest
Alternative 1 – Continuous Mobile Calibration
04/10/2012 35© 2012, VITO NV
» Deployment of a subset of Sensor Boxes close to portable
and “TRUSTABLE” device in a mobile contest
» Classification Problem
» Target: [BC, UFP]
Alternative 2 – Discrete Mobile Calibration
1±0.5E5
2±0.5E5
3±0.5E5
3±0.5E5
04/10/2012 36© 2012, VITO NV
Acknowledgments
» The EveryAware project is funded under FP7 -
Information Society Technologies, IST - FET Open Scheme
» The IDEA project is funded by the Flemish Agency for
Innovation through Science and Technology
Contact