CSR_Module5_Green Earth Initiative, Tree Planting Day
What does the future hold for low cost air pollution sensors? - Dr Pete Edwards
1. Dr Pete Edwards
Wolfson Atmospheric Chemistry Laboratories, University of York, UK.
What does the future hold for low cost
air pollution sensors?
2. The need for atmospheric chemical measurements
o Monitor regulation compliance
o Check policy efficacy
o Improve our understanding
o Measure personal exposure?
3. Metal oxide
~ €5-10
Electrochemical
~ €50-100
Regulation grade instrument
~ €10,000 – €100,000
Relies on assumption that the
sensor data is fit for purpose.
Micro-optical
> €100-1000
A revolution in air pollution measurement?
4. “Hype Cycle”
model used by
Gartner since
1995 and the
“Technology
Adoption
Lifecycle” model
popularized by
Rogers and
Moore
Hype Cycle and Technology
Adoption Lifecycle
A revolution in air pollution measurement?
7. Key challenges
Variability - Market
- Evaluation
- Sensor to sensor
14
Alphasense reported 𝑅 ~ 1 for hourly data. High values of 𝑅 were also
reported for the SSys AirSensEUR (v.2), when calibrating CO minute data41
(Figure A2). Other LCSs reporting values of 𝑅2
within the range 0.7 - 1.0 for
hourly data consisted of the MICS-4515 by and SGX Sensortech60
, the Smart
Citizen Kit by Acrobotic3
and the RAMP93
Figure 3. Mean 𝑅2 for obtained from the calibration of sensor systems against reference
measurements.
Taken from EC Joint Research Centre report
“Review of sensors for air quality monitoring” 2019.
8. Key challenges
Variability - Market
- Sensor to sensor
- Evaluation
Complex interferences from other pollutants and
physical parameters
Factory calibrations not applicable to real world
14
Alphasense reported 𝑅 ~ 1 for hourly data. High values of 𝑅 were also
reported for the SSys AirSensEUR (v.2), when calibrating CO minute data41
(Figure A2). Other LCSs reporting values of 𝑅2
within the range 0.7 - 1.0 for
hourly data consisted of the MICS-4515 by and SGX Sensortech60
, the Smart
Citizen Kit by Acrobotic3
and the RAMP93
Figure 3. Mean 𝑅2 for obtained from the calibration of sensor systems against reference
measurements.
Taken from EC Joint Research Centre report
“Review of sensors for air quality monitoring” 2019.
9. Key challenges
Variability - Market
- Sensor to sensor
- Evaluation
Complex interferences from other pollutants and
physical parameters
Factory calibrations not applicable to real world
Data quality MUST fit with application
New methods needed to exploit the potential of
these technologies!
14
Alphasense reported 𝑅 ~ 1 for hourly data. High values of 𝑅 were also
reported for the SSys AirSensEUR (v.2), when calibrating CO minute data41
(Figure A2). Other LCSs reporting values of 𝑅2
within the range 0.7 - 1.0 for
hourly data consisted of the MICS-4515 by and SGX Sensortech60
, the Smart
Citizen Kit by Acrobotic3
and the RAMP93
Figure 3. Mean 𝑅2 for obtained from the calibration of sensor systems against reference
measurements.
Taken from EC Joint Research Centre report
“Review of sensors for air quality monitoring” 2019.
10. Clusters of ‘identical’ sensors average out sensor to
sensor variability.
Use combination of ‘average sensors’ with supervised
learning algorithms to correct for interferences.
Sensor to sensor variability
The York Box of Clustered Sensors (BOCS)
11. Factory sensor calibrations significantly over predict actual NO2
Factory calibrations not applicable to real world
Beijing 2017
12. In-field co-location calibration improves agreement, but still drifts after a week. – Expensive!
Beijing 2017
Factory calibrations not applicable to real world
13. Complex interferences
The large number of complex non-linear interferences makes traditional univariant
corrections difficult.
Calibration methods needed that provide robust calibrations across multivariate
chemical / physical space.
Working electrode responses (in mV ppb-1
of co-pollutant) induced by the presentation of co-pollutants
in zero air across five electrochemical sensors
Co-pollutants
Potentially significant interference
CompoundSensor
14. ure 5: Validation results using multivariate linear regression (left) and k Nearest Neighbors
ression (right). Data are shown as the SO2 measurement by a single sensor (SO2-02) vs. the reference
ech. Discuss., https://doi.org/10.5194/amt-2017-296
er review for journal Atmos. Meas. Tech.
ed: 15 August 2017
17. CC BY 4.0 License.
Used electrocheical sensors to measure SO2
on Hawaii
Found interferences from temperature etc.
Had most impact at low SO2
Compared calibrations using linear regression
and k nearest neighbours
Can a potential solution lay in supervised machine learning type approaches to model
sensor response?
Complex interferences
Calibration and assessment of electrochemical air
quality sensors by co-location with reference-grade
instruments
David H. Hagan1
, Gabriel Issacman-Vanwertz1,2
, Jonathan P. Franklin1
, Lisa M. M. Wallace3
,5
Benjamin D. Kocar1
, Colette L. Heald1,4
, Jesse H. Kroll1,5
1
Department of Civil and Environmental Engineering, Massachusetts Institute of Technology, Cambridge, 02139,
USA
2
Department of Civil and Environmental Engineering, Virginia Tech, Blacksburg, 24061, USA
3
Air Surveillance and Analysis Section, Hawai’i State Department of Health, Hilo, 96720, USA10
4
Atmos. Meas. Tech. Discuss., https://doi.org/10.5194/amt-2017-296
Manuscript under review for journal Atmos. Meas. Tech.
Discussion started: 15 August 2017
c Author(s) 2017. CC BY 4.0 License.
15. 3725
3724 A. Bi
mate
data,
In
dicat
NO2
algor
tants
of th
plete
EC s
phas
to C
Kim
was s
(201
vide
and f
Performance of NO, NO2 low cost sensors and three calibration
approaches within a real world application
Alessandro Bigi1, Michael Mueller2, Stuart K. Grange3, Grazia Ghermandi1, and Christoph Hueglin2
1“Enzo Ferrari” Department of Engineering, University of Modena and Reggio Emilia, Modena, Italy
2Empa, Swiss Federal Laboratories for Materials Science and Technology, Duebendorf, Switzerland
3Wolfson Atmospheric Chemistry Laboratory, University of York, York, UK
Correspondence: Alessandro Bigi (alessandro.bigi@unimore.it)
Received: 16 February 2018 – Discussion started: 12 March 2018
Revised: 29 May 2018 – Accepted: 12 June 2018 – Published: 26 June 2018
Abstract. Low cost sensors for measuring atmospheric pol-
lutants are experiencing an increase in popularity worldwide
among practitioners, academia and environmental agencies,
and a large amount of data by these devices are being deliv-
ered to the public. Notwithstanding their behaviour, perfor-
mance and reliability are not yet fully investigated and under-
stood. In the present study we investigate the medium term
performance of a set of NO and NO2 electrochemical sensors
in Switzerland using three different regression algorithms
within a field calibration approach. In order to mimic a re-
alistic application of these devices, the sensors were initially
co-located at a rural regulatory monitoring site for a 4-month
calibration period, and subsequently deployed for 4 months
at two distant regulatory urban sites in traffic and urban back-
ground conditions, where the performance of the calibration
algorithms was explored. The applied algorithms were Mul-
tivariate Linear Regression, Support Vector Regression and
Random Forest; these were tested, along with the sensors, in
terms of generalisability, selectivity, drift, uncertainty, bias,
noise and suitability for spatial mapping intra-urban pollu-
tion gradients with hourly resolution. Results from the de-
ployment at the urban sites show a better performance of the
non-linear algorithms (Support Vector Regression and Ran-
dom Forest) achieving RMSE < 5 ppb, R2 between 0.74 and
0.95 and MAE between 2 and 4 ppb. The combined use of
both NO and NO2 sensor output in the estimate of each pol-
lutant showed some contribution by NO sensor to NO2 es-
timate and vice-versa. All algorithms exhibited a drift rang-
ing between 5 and 10 ppb for Random Forest and 15 ppb for
Multivariate Linear Regression at the end of the deployment.
The lowest concentration correctly estimated, with a 25 %
relative expanded uncertainty, resulted in ca. 15–20 ppb and
was provided by the non-linear algorithms. As an assessment
for the suitability of the tested sensors for a targeted applica-
tion, the probability of resolving hourly concentration differ-
ence in cities was investigated. It was found that NO concen-
tration differences of 5–10 ppb (8–10 for NO2) can reliably
be detected (90 % confidence), depending on the air pollu-
tion level. The findings of this study, although derived from a
specific sensor type and sensor model, are based on a flexible
methodology and have extensive potential for exploring the
performance of other low cost sensors, that are different in
their target pollutant and sensing technology.
1 Introduction
Air quality assessment for regulatory purposes is addressed
by means of monitoring stations following a strict QA/QC
protocol in order to deliver measurements having an uncer-
tainty within a specific range that is appropriate for the pur-
pose (2008/50/EC, Council of Europe, 2008). The costs as-
sociated to these monitoring sites led to a reconfiguration of
regulatory air quality networks across Europe over the last
decade, resulting in improved but still spatially sparse regu-
latory air quality networks over the continent. Although this
trend towards optimisation is coherent with main regulatory
needs, it is not consistent with the increasing demand for
spatio-temporal air quality information in urban areas, where
largest part of worldwide population lives (United Nations,
2015). Up to now, two of the most promising approaches for
estimating air quality conditions in complex environments
Published by Copernicus Publications on behalf of the European Geosciences Union.
Complex interferences
Tested a range of algorithms to calibarte NO
and NO2 sensors in Switzerland.
Significant improvements using non-linear
methods, but calibrations did drift over time.
Identifies key interferants
meseries of 1 week rolling RMSE for 10 min data of Figure 8. Time series of mean daily residuals for NO and NO2 es-
16. Supervised learning algorithms able to use all the sensor measurements to correct
for interferences, providing a more robust calibration.
Beijing 2017
Complex interferences
18. Sensor accuracy significantly improved by in-
field calibration (costly!!)
ML algorithms can improve calibrations – BUT
require large amounts of data across all
sampled conditions to be robust long-term
Complex interferences
19. Sensor accuracy significantly improved by in-
field calibration (costly!!)
ML algorithms can improve calibrations – BUT
require large amounts of data across all
sampled conditions to be robust long-term
Air Quality Sensors and Data Adjustment Algorithms: When Is It No
Longer a Measurement?
Gayle S. W. Hagler,*,†
Ronald Williams,†
Vasileios Papapostolou,‡
and Andrea Polidori‡
†
United States Environmental Protection Agency, Office of Research and Development, National Exposure Research Laboratory,
Research Triangle Park, North Carolina 27709, United States
‡
South Coast Air Quality Management District, Diamond Bar, California 91765, United States
sensor device is relocated to another environment for ongoing
use and the correction algorithm is applied, based upon the
presumption that the ongoing sampling conditions are within
range of the calibration period. In some approaches, sensor data
at one location are adjusted based upon measurements in other
places, assuming there is homogeneity in air pollution
concentrations over a specific geographic area and time
frame;5
for example, this approach appears to be supported
via commercially available software (e.g., Advanced Normal-
ization Tool for AirVision; http://agilaire.com/pdfs/ANT.pdf).
These emerging strategies raise a number of questions for
Viewpoint
pubs.acs.org/estCite This: Environ. Sci. Technol. 2018, 52, 5530−5531
09:43:02(UTC).
atelysharepublishedarticles.
Complex interferences
20. Low cost sensor networks are a VERY exciting
opportunity!
Regular in-field calibration of every node in a sensor
network not practical
New methods needed for both validation and use of
data
(e.g. using nearby reference monitors under certain
conditions to check sensor to sensor variability)
Trials are costly but progress is being made
The power of the network
1938 J. Kim et al.: BErkeley Atmospheric CO2 Observation Ne
30 km
10 km
Figure 1. Map of San Francisco Bay Area showing current BEACO2N node sites (red), BAAQMD reference sites with O3 measu
(blue), and the BAAQMD Bodega Bay regional greenhouse gas background site (orange). The sites used in this analysis are marked in
on the detailed panel.
21. What does the future hold?
• Low cost sensors are an exciting opportunity.
• Issues such as sensor-to-sensor variability and cross-interferences need to be
considered.
• Data quality and calibration are key!!!
• Large amount of market and performance evaluation variability
• Physical and data solutions to key challenges look promising
• Cost effective methods to calibrate networks are also being developed
• New ways of using data from low cost devices, acknowledging weaknesses as well as
strengths, still needed to truly realise potential
22. Quantification of Utility of Atmospheric Network
Technologies (QUANT)
• Real-world open and fully-traceable assessment of low-cost sensors and sensor
networks, including calibration methods, at 4 sites across 3 UK cities over 2 years.
• Enhance the value of low-cost sensor data for UK air pollution challenges through
the development of novel methods to extract new information on key pollutants,
significantly enhancing the use of sensors for exposure and policy intervention
studies.