data quality evaluation of low-cost pm10 air quality sensors · aerasense nanomonitor pnmt 1000. o....
TRANSCRIPT
DATA QUALITY EVALUATION OF LOW-COSTPM10 AIR QUALITY SENSORSV.M. VAN ZOEST, N.A.S. HAMM, A. STEIN (ITC, UNIVERSITY OF TWENTE)G. HOEK (INSTITUTE OF RISK ASSESSMENT SCIENCE (IRAS), UTRECHT UNIVERSITY)R. OTJES (ENERGY RESEARCH CENTRE OF THE NETHERLANDS (ECN))
PROJECT DAMAST: “DEVELOPMENT OF AN AUTOMATIC SYSTEM FOR MAPPING AIR QUALITY RISKS IN SPACE AND TIME”
THE AIR QUALITY SENSOR NETWORK
Geospatial Sensor Webs Conference, 29-31 August 2016 2
www.aireas.com
DATA
Reference data:
• RIVM LML data (national ambient air quality monitoring network)
• 2 authorative sensors (BAMs) in Eindhoven, 3 in surrounding area
Geospatial Sensor Webs Conference, 29-31 August 2016 3
Variable Units Instrument
PM10 μg m-3 Shinyei PPD42 ECN revised
PM2.5 μg m-3 Shinyei PPD42 ECN revised
PM1 μg m-3 Shinyei PPD42 ECN revised
UFP # cm-3 Aerasense NanoMonitor PNMT 1000
O3 μg m-3 E2V MICS 2610
NO2 μg m-3 Citytech Sensoric NO2 3E50 ECN revised
Temperature °C Sensirion SHT75
Relative humidity % Sensirion SHT75
RESEARCH QUESTIONS
Data quality of low-cost sensors largely unknown. The followingquestions arise:
What is the precision of the sensors?
How do the low-cost sensors perform compared to the authorativeLML monitors (BAMs) in Eindhoven?
Geospatial Sensor Webs Conference, 29-31 August 2016 4
RESEARCH QUESTIONS
Data quality of low-cost sensors largely unknown. The followingquestions arise:
What is the precision of the sensors?
How do the low-cost sensors perform compared to the authorativeLML monitors (BAMs) in Eindhoven?
Geospatial Sensor Webs Conference, 29-31 August 2016 5
SENSOR PRECISION: METHODS
Co-locating sets of 12 airboxes at location RIVM LML monitor (BAM)
3 sets of airboxes, each 2 sessions of ~1 week
Calculate standard deviation (SD) and coefficient of variation (CV) across each set of sensors for each hour
𝐶𝐶𝐶𝐶 = 𝑆𝑆𝑆𝑆𝑚𝑚
× 100%
Geospatial Sensor Webs Conference, 29-31 August 2016 6
Calculate aggregatedstatistics on mean, SD andCV per session
Compare variability betweensensors when co-located andwhen at their usual locations
SENSOR PRECISION: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 7
Series Session Mean of mean (µg m-3)
Mean SD (µg m-3)
Median SD(µg m-3)
Mean CV (%)
Series 1 Session 1 9.7 1.3 1.2 13.1
Session 2 10.0 1.3 1.3 13.3
Series 2 Session 1 15.9 2.7 2.6 16.4
Session 2 12.7 1.9 1.6 14.0
Series 3 Session 1 13.2 2.7 2.3 18.8
Session 2 15.2 2.8 2.4 18.0
Precision of PM10 sensors August-December 2015
SENSOR PRECISION: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 8
SENSOR PRECISION: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 9
SENSOR PRECISION: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 10
SENSOR PRECISION: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 11
Comparison precision and spatial variation airboxes 2015 (hourly PM10 concentration values)
Serie & session
Period Mean of mean (µg m-3)
Mean SD(µg m-3)
Median SD(µg m-3)
Mean CV(%)
Serie 1 validation
15/08/2015 – 18/08/2015 +25/09/2015 – 04/10/2015 10.1 1.4 1.3 13.4
Serie 1 regular
01/01/2015 – 13/08/2015 +06/10/2015 – 31/12/2015 15.3 2.8 2.3 18.6
Serie 2 validation
08/10/2015 – 11/10/2015 +04/11/2015 – 12/11/2015 14.0 2.2 2.1 15.0
Serie 2 regular
01/01/2015 – 06/10/2015 +14/11/2015 – 31/12/2015 15.9 3.0 2.6 18.6
Serie 3 validation
17/11/2015 – 30/11/2015 +19/12/2015 – 27/12/2015 13.3 2.6 2.2 18.3
Serie 3 regular
01/01/2015 – 15/11/2015 +29/12/2015 – 31/12/2015 15.2 3.0 2.6 19.6
Regular Validation Maintenance Validation Regular
RESEARCH QUESTIONS
Data quality of low-cost sensors largely unknown. The followingquestions arise:
What is the precision of the sensors?
How do the low-cost sensors perform compared to the authorativeLML monitors (BAMs) in Eindhoven?
Geospatial Sensor Webs Conference, 29-31 August 2016 12
SENSOR ACCURACY: METHODS
2 airboxes permanently co-located with authorative sensors (BAMs of RIVM LML network)
Derive time plots, scatterplots, statistical indicators using dailyaverage concentrations over full year 2015
Calculate sensor uncertainty compared to BAM using Guide to theDemonstration of Equivalence
Geospatial Sensor Webs Conference, 29-31 August 2016 13
SENSOR ACCURACY: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 14
Con
cent
ratio
n(µ
g/m
3 )
1/1/2015 31/12/2015Date
SENSOR ACCURACY: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 15
SENSOR ACCURACY: RESULTS
Geospatial Sensor Webs Conference, 29-31 August 2016 16
Con
cent
ratio
n(µ
g/m
3 )
1/1/2015 31/12/2015Date
SENSOR UNCERTAINTY: METHODS
1. Uncertainty between airboxes
2. Uncertainty between LML monitoring stations
3. Calibration airbox to LML monitoring station
4. Uncertainty airbox (calibrated values) in relation to LML station (equivalence test)
5. Derive uncertainty at threshold concentration of 50 µg m-3
(uncertainty should be max. 25%)
17
EC WORKING GROUP ON GDE 2010. Guide to the Demonstration of Equivalence of Ambient Air Monitoring Methods. European Commission.
Geospatial Sensor Webs Conference, 29-31 August 2016
SENSOR UNCERTAINTY: RESULTS
18Geospatial Sensor Webs Conference, 29-31 August 2016
Uncertainty at 50 µg/m3: 27.2%
INFLUENCE OF WIND DIRECTION
19Geospatial Sensor Webs Conference, 29-31 August 2016
Con
cent
ratio
ndi
ffere
nce
(µg/
m3 )
Wind direction
CONCLUSION
Spatial variability is larger than the noise of the airbox sensors
Moderate correlation between airbox and LML monitor based on 24h average values
Uncertainty of airbox compared to LML just above threshold valuefor equivalence
Airbox underestimates values of LML, especially whenconcentrations peak (airbox does not detect particles <0.3µm)
Geospatial Sensor Webs Conference, 29-31 August 2016 20
CONCLUSION
Data quality of low-cost sensors largely influenced by:
Outliers
Erroneous values
Pollutant composition
There is a need for:
Automatic data quality evaluation
Outlier detection
Provision of data quality information to the users
Standard calibration methods on finer temporal resolutions
Geospatial Sensor Webs Conference, 29-31 August 2016 21
THANK YOU FOR YOUR ATTENTIONANY QUESTIONS?