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

» jan.theunis@vito.be