*>EPA
United States
Environmental Protection
Agency
Spatial and Temporal Trends of Air
Pollutants in the South Coast Basin
¦laiftrte
Using Low Cost Sensors
ii
| 409: Valley View Elementary School
| 406: Saul Martinez Elementary School)
Office of Research and Development
National Exposure Research Laboratory
-------
EPA/600/R-17/463
January 2018
Spatial and Temporal Trends of Air
Pollutants in the South Coast Basin
Using Low Cost Sensors
Ron Williams and Rachelle Duvall
U.S. Environmental Protection Agency
Office of Research and Development
National Exposure Research Laboratory
Research Triangle Park, NC 27711
Dena Vallano
U.S. Environmental Protection Agency, Region 9
Air Quality Analysis Office
San Francisco, CA 95105
Andrea Polidori, Vasileios Papapostolou, Brandon Feenstra, Hang Zhang
South Coast Air Quality Management District Science and Technology Advancement Office
Air Quality Sensor Performance Evaluation Center (AQ-SPEC)
Diamond Bar, CA 91765
Sam Garvey
Jacobs Technology Inc.
Tullahoma, TN 37388
-------
Disclaimer
This technical report presents the results of work performed by the South Coast Air Quality Management
District's AQ-SPEC Laboratory under contract EP-16-W-000117 and Jacobs Technology, Inc. under contract
EP-C-15-008 for the National Exposure Research Laboratory, U.S. Environmental Protection Agency (U.S.
EPA), Research Triangle Park, NC. It has been reviewed by the U.S. EPA and approved for publication.
Mention of trade names or commercial products does not constitute endorsement or recommendation for
use.
Abstract
The emergence of small, portable, low-cost air sensors has encouraged a shift toward their use and away from
traditional approaches to monitoring air quality. The U.S. Environmental Protection Agency (U.S. EPA), in
collaboration with the South Coast Air Quality Management District's (SCAQMD) Air Quality Sensor
Performance Evaluation Center (AQ-SPEC), deployed custom-built sensor devices (pods) measuring fine
particulate matter (PM2.5), ozone (03), relative humidity, and temperature at nine locations throughout
southern California from January 2017 to April 2017 to evaluate their performance under "real-life"
conditions. Prior to the deployment, these pods were evaluated within the AQ-SPEC program both in the
field and in the laboratory. Southern California is an ideal testing location for air quality sensor technology,
as it often experiences elevated air pollutant levels resulting from gasoline and diesel engines, marine
ports, and various other industries. The South Coast Air Basin's particular meteorology (frequent sunny
days and little precipitation) and geography also contribute to the elevated pollution levels in the region.
The goal of this project was to characterize the performance of these newly developed pods and
better understand their potential applications for community monitoring. This report summarizes the AQ-
SPEC field and laboratory performance evaluations of the Citizen Science Air Monitor (CSAM) sensor pods
designed and developed by the EPA. In addition, this document summarizes the spatial and temporal
variability of the PM2.5 and 03 measurements collected during the field deployment of the CSAM pods
at nine monitoring locations covering approximately a 200 km2 area in southern California.
-------
Table of Contents
ABSTRACT Ill
TABLE OF CONTENTS IV
ACRONYMS AND ABBREVIATIONS XII
EXECUTIVE SUMMARY XIV
Background xiv
Study Objectives xiv
Study Approach xiv
Conclusions xvi
INTRODUCTION 1
1. CO-LOCATION TESTING 1
1.1. Methods 1
1.2. Results 3
1.2.2. Ambient Ozone 13
1.2.3. AmbientTemperature 20
1.2.4. ambient Relative Humidity 25
1.3. QA Summary for Co-location Testing 31
2. LABORATORY CHAMBER EVALUATION 33
2.1. Methods 33
2.1.1. PM2.5 TESTING PROCEDURE 35
2.1.2. Ozone testing procedure 36
2.1.3. CSAM Evaluation Parameters 37
2.2. Results 39
2.2.1. Laboratory pm2.5 39
2.2.2. Laboratory Ozone 42
2.3. QA SUMMARY FOR LABORATORY TESTING 45
3. FIELD DEPLOYMENT 46
3.1. Methods 46
3.2. Results 51
3.2.1. Ambient PM25 52
iv
-------
3.2.2. Ambient Ozone 58
3.2.3. Temperature and RH influences 65
3.3. QA SUMMARY FOR FIELD DEPLOYMENT 68
CONCLUSIONS 70
REFERENCES 71
v
-------
List of Tables
Table A.l. Sensor components in each CSAM Sensor Pod developed by EPA xv
Table 1.1a. Descriptive statistics for PM2.5 sensors in the CSAMs and the FEM GRIMM reference instrument (5-
MINUTE AVERAGE) 4
Table 1.1b. Correlation statistics for PM2.5 sensors in the CSAMs against the FEM GRIMM reference
INSTRUMENT (5-MINUTE AVERAGE), [FEM= (SLOPE*SENSOR READING) + INTERCEPT] 4
Table 1.2. Correlation Coefficient (R ) matrix for the 5-minute average PM2.5 mass concentrations measured
BY THE FEM AND CSAM UNITS 6
Table 1.3. Correlation Coefficient (R ) matrix for the 1-hour average PM2.5 mass concentrations measured
BY THE FEM AND CSAM UNITS 8
Table 1.4. Correlation Coefficient (R ) matrix for the 24-hour average PM2.5 mass concentrations measured
BYTHE FEM AND CSAM UNITS 9
Table 1.5. Correlation Coefficient (R ) matrix for the 1-hour average PM2.5 mass concentrations measured
BYTHE FEM AND CSAM UNITS 11
Table 1.6. Correlation Coefficient (R ) matrix for the 1-hour average PM2.5 mass concentrations measured
BYTHE FEM AND CSAM UNITS 12
Table 1.7a. Descriptive Statistics for ozone from the CSAMs and the FEM instrument 13
Table 1.7b. Correlation statistics for ozone from the CSAMS against the FEM instrument (5- minute average)
[FEM = (SLOPE*SENSOR READING) + INTERCEPT] 13
Table 1.8. Correlation Coefficient (R2) matrix for the 5-minute average ozone concentrations measured by
THE FEM AND CSAM UNITS 15
Table 1.9. Correlation coefficient (R2) matrix for the 1-hour average ozone concentrations measured bythe
FEM AND CSAM UNITS 17
Table 1.10. Correlation coefficient (R2) matrix for the 8-hour average ozone concentrations measured by
THE FEM AND CSAM UNITS 18
Table 1.11. Correlation coefficient (R2) matrix for the 24-hour average ozone concentrations measured by
THE FEM AND CSAM UNITS 20
Table 1.12. Descriptive statistics for temperature data in the nine CSAMs and RIVR station 20
Table 1.13. Correlation coefficient (R2) matrix for the 5-minute average temperature values measured by the
WEATHER STATION SENSOR AND CSAM UNITS 22
Table 1.14. Correlation coefficient (R2) matrix for the 1-hour average temperature values measured by the
RIVR WEATHER STATION SENSOR AND CSAM UNITS 24
Table 1.15. Correlation coefficient (R2) matrix for the 24-hour average temperature measurements taken by
THE RIVR WEATHERSTATION SENSOR AND CSAM UNITS 25
vi
-------
Table 1.16. Descriptive statistics for the Relative Humidity sensors for the CSAM units and the RIVR station 26
Table 1.17. Correlation Coefficient (R2) matrix for the 5-minute average relative humidity values measured by
THE WEATHER STATION SENSOR AND CSAM UNITS 28
Table 1.18. Correlation Coefficient (R2) matrix for the 1-hour average RH values measured by the RIVR
WEATHER STATION SENSOR AND CSAM UNITS 29
Table 1.19. Correlation coefficient (R2) matrix for the 24-hour average RH values measured by the RIVR
WEATHER STATION SENSOR AND CSAM UNITS 31
Table 2.1. Four representative T-RH combinations 36
Table 2.2. CSAM PM2.5 Accuracy 40
Table 2.3. CSAM PM2.5 precision under extreme T and RH conditions 40
Table 2.4. CSAM Ozone Accuracy 42
Table 2.5. CSAM Ozone Precision under extreme T and RH conditions 43
Table 3.1. CSAM Deployment Locations 47
Table 3.2 Deployment Location and Dates 49
Table 3.3. Percent Data Recovery 52
Table 3.4. CSAM PM2.5 Summary Statistics 53
Table 3.5. FEM MetoneBAM PM2.5 Summary Statistics 53
Table 3.6. CSAM Ozone Summary Statistics 58
Table 3.7. FEM Air Monitoring Station Ozone Summary Statistics 59
Table 3.8. FRM and CSAM #410 ozone sensor regression parameters by month 64
List of Figures
Figure A.l. Sensor pod assembly xv
Figure A.2. Fully assembled pod with solar panel, battery cell, and tripod xv
Figure 1.1. Co-location set-up for CSAM units #401 - #407 at Riverside-Rubidoux AMS 2
Figure 1.2. Co-location set-up for CSAM units #408 - #410 at Riverside-Rubidoux AMS 2
Figure 1.3. Intra-model variability for nine of the ten CSAM PM2.5 sensors tested. Vertical bars represent the
STANDARD DEVIATION FOR EACH MEAN VALUE 4
Figure 1.4. Correlation coefficient (R ) plot for the 1-hour average PM2.5 measurements by the FEM GRIMM
AND FEM BAM UNITS (1-HOUR AVERAGE) 5
Figure 1.5. Time-series plot of PM2.5 measurements from the FEM GRIMM vs FEM BAM units (1-hour average)
5
Figure 1.6. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM instrument (5-
MINUTE AVERAGE) 6
vii
-------
Figure 1.7. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM instrument
(5-minute average) 6
Figure 1.8. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM instrument
(1-HOUR AVERAGE) 7
Figure 1.9. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM instrument
(1-HOUR AVERAGE) 7
Figure 1.10. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM instrument
(24-hour average) 8
Figure 1.12. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM instrument
(1-HOUR AVERAGE) 10
Figure 1.13. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM instrument
(1-HOUR AVERAGE) 10
Figure 1.14. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM instrument
(24-hour average) 11
Figure 1.15. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM instrument
(1-HOUR AVERAGE) 12
Figure 1.17. Time-series plot of ozone measurements from units #401 through #404 and the FEM 14
INSTRUMENT (5-MINUTE AVERAGE) 14
Figure 1.18. Time-series plot of ozone measurements from units #406 through #410 and the FEM instrument
(5-minute average) 15
Figure 1.19. Time-series plot of ozone measurements from units #401 through #404 and the FEM instrument
(1-HOUR AVERAGE) 16
Figure 1.20. Time-series plot of ozone measurements from units #406 through #410 and the FEM instrument
(1-HOUR AVERAGE) 16
Figure 1.21. Time-series plot of ozone measurements from units #401 through #404 and the FEM instrument
(8-hour average) 17
Figure 1.22. Time-series plot of ozone measurements from units #406 through #410 and the FEM instrument
(8-hour average) 18
Figure 1.23. Time-series plot of ozone measurements from units #401 through #404 and the FEM instrument
(24-hour average) 19
Figure 1.24. Time-series plot of ozone measurements from units #406 through #410 and the FEM instrument
(24-hour average) 19
Figure 1.25. Intra-model variability for nine CSAM temperaturesensors 21
Figure 1.26. Time-series plot of temperature measurements from units #401 through #404 and the weather
STATION SENSOR (5-MINUTE AVERAGE) 21
Figure 1.28. Time-series plot of temperature measurements from units #401 through #406 and the weather
STATION SENSOR (1-HOUR AVERAGE) 23
Figure 1.29. Time-series plot of temperature measurements from units #406 through #410 and the weather
STATION SENSOR (1-HOUR AVERAGE) 23
viii
-------
Figure 1.30. Time-series plot of temperature measurements from units #401 through #404 and the R I V R
WEATHERSTATION SENSOR (24-HOUR AVERAGE) 24
Figure 1.31. Time-series plot of temperature measurements from units #406 through #410 and the R I V R
WEATHERSTATION SENSOR (24-HOUR AVERAGE) 25
Figure 1.32. Intra-model variability for the nine CSAM RH sensors evaluated in this study 26
Figure 1.33. Time-series plot of RH measurements from units #401 through #404 and the R I V R weather
STATION SENSOR (5-MINUTE AVERAGE) 27
Figure 1.34. Time-series plot of RH measurements from units #406 through #410 and the RIVR weather
STATION SENSOR (5-MINUTE AVERAGE) 27
Figure 1.35. Time-series plot of RH measurements from units #401 through #404 and the weather station
SENSOR (1-HOUR AVERAGE) 28
Figure 1.36. Time-series plot of RH measurements from units #406 through #410 and the weather station
SENSOR (1-HOUR AVERAGE) 29
Figure 1.37. Time-series plot of RH measurements from units #401 through #404 and the RIVR weather
STATION SENSOR (24-HOUR AVERAGE) 30
Figure 1.38. Time-series plot of RH measurements from units #406 through #410 and the RIVR weather
STATION SENSOR (24-HOUR AVERAGE) 30
Figure 2.1. Schematic of AQ-SPEC's environmental chamber 34
Figure 2.2. Two CSAM units installed on the inner chamber base 35
Figure 2.3. PM2.5 mass concentrations as measured by the CSAM and the reference instruments used for these
TESTS (FEM GRIMM AND APS) 39
Figure 2.4. Effect ofT-RH on CSAM PM25 Performance 41
Figure 2.5. (a) Unit #406 vs unit #409; (b-d) intra-model variability at low, medium, and high PM2.5 conc 41
Figure 2.6. CSAM vs FEM ozone concentration ramping experiment (20 °C, 40% RH) 42
Figure 2.7. Effect of T-RH on CSAM ozone performance 43
Figure 2.8. (a) Unit #406 vs unit #409; (b-d) intra-model variability at low, medium, and high ozone
CONCENTRATIONS 44
Figure 2.9. Effect of N02 interferent 44
Figure 3.1. Sensor Pod Placement in the South Coast Air Basin 48
Figure 3.2. CSAM Pod #401 at Hudson AMS 49
Figures 3.3. CSAM Pod #402 at710 NR 50
Figures 3.4. CSAM Pod #404 at Ventura Transfer Company 50
Figures 3.5. CSAM Pod #406 at Saul Martinez Elementary School 50
Figures 3.6. CSAM Pod #407 at Jurupa Area Recreation and Parks District 50
Figures 3.7. CSAM Pod #408 at Mira Loma AMS 51
Figures 3.8. CSAM Pod #409 at Valley View Elementary School 51
Figure 3.9. CSAM Pod #410 at Indio AMS 51
ix
-------
Figure 3.10 Long Beach PM2.5 54
Figure 3.11. Hudson AMS time series 54
Figure 3.12. Hudson AMS and 710 NR AMS 55
Figure 3.13. Jurupa Valley PM2.5 24-hour time series 55
Figure 3.14. Rubidoux FEM BAM and CSAM PM2.5 time series 56
Figure 3.15. Mira Loma FEM BAM and JARPD CSAM 24-hour PM2.5 Data 56
Figure 3.16. Rubidoux Co-located CSAM time series 57
Figure 3.17. Rubidoux Co-Location Scatterplots Fulltime Period (left) and Filtered for January 11, 2017 to
February 11, 2017 (right) 57
Figure 3.18. Time series of Rubidoux Co-located CSAM Pods #405 and #407 57
Figure 3.19. Time series of CSAM Pod #410 in Coachella Valley 58
Figure 3.20. Hudson FEM and CSAM #401 time series 59
Figure 3.21. Hudson FEM and CSAM #401 scatterplot 60
Figure 3.22. Long Beach Ozone 24-hour time series 60
Figure 3.23. Long Beach Ozone 8-hour time series 60
Figure 3.24. Jurupa Valley Ozone Concentrations, 24-hour Mean 61
Figure 3.25. Jurupa Valley Ozone Time series, 8-hour average 61
Figure 3.26. Rubidoux Co-Location Ozone time series, 8-hour average 62
Figure 3.27. Rubidoux Co-Location Correlation Plots: a. Full time period, b. Filtered time to include January
1, 2017 to February 26, 2017 62
Figure 3.28. Rubidoux FRM ozone and CSAM #403 and #405 comparison 62
Figure 3.29. Coachella Valley Ozone, 24-hour Mean 63
Figure 3.30. Coachella Valley Ozone, 8-hour rolling average 63
Figure 3.31. Indio 63
Figure 3.32. Indio AMS FRM and CSAM #410 scatterplot 64
Figure 3.33. Time series of FRM and CSAM Pods #406, #409 and #410 (24-hour average) 65
Figure 3.34. Indio 410 Temperature vs. Ozone 66
Figure 3.36. Indio 410 Dew Point and Ozone 66
Figure 3.37 Hudson 401Temp and PM 66
Figure 3.38. Hudson 401 Humidity and PM 66
Figure 3.39. Hudson 401 Dew Point and PM 67
Figure3.40. JARPD407Temp and PM 67
Figure 3.41. JARPD 407 Humidity and PM 67
Figure 3.42. JARPD Dew Point and PM 67
x
-------
Figure3.43. Indio410Temp and PM 68
Figure 3.45. Indio 410 Dew Point and PM 68
xi
-------
Acronyms and Abbreviations
AMS
AM
APS
AQ-SPEC
BAM
BAT
°C
COTS
CSAM
DP
EJ
EPA
°F
FEM
FRM
hr
JARPD
m
Max
Hg/m3
min
Min
mL
NAAQS
NO
nm
N02
03
OPC
ORD
PM
PM1.0
PM2.5
PM10
PMC
Ppb
QAPP
QA/QC
R2
RH
RIVR
ROP
RTP
SCAB
SCAQMD
Air Monitoring Station
Atmospheric Measurements
Aerodynamic Particle Sizer
Air Quality - Sensor Performance Evaluation Center
Beta Attenuation Monitor
Best Available Technology
Degrees Celsius
Commercial-off-the-shelf
Citizen Science Air Monitor
Dew point
Environmental Justice
Environmental Protection Agency
Degrees Fahrenheit
Federal Equivalent Method
Federal Reference Method
Hour
Jurupa Area Recreation and Parks District
meter
maximum
Microgram per cubic meter
Minute
Minimum
Milliliter
National Ambient Air Quality Standards
Nitric oxide
nanometer
Nitrogen dioxide
Ozone
Optical Particle Counter
Office of Research and Development
Particulate matter
Particles smaller than 1 nm in aerodynamic diameter
Particles smaller than 2.5 nm in aerodynamic diameter
Particles smaller than 10 nm in aerodynamic diameter
Coarse Particulate matter
Parts per billion
Quality Assurance Project Plan
Quality Assurance/Quality Control
Correlation coefficient
Relative humidity
Rubidoux Air Monitoring Station
Research Operating Protocol
Research Triangle Park
South Coast Air Basin
South Coast Air Quality Management District
XII
-------
SE
Standard Error
SOP
Standard Operating Procedure
S02
Sulphur Dioxide
SSAB
Salton Sea Air Basin
STDEV
Standard Deviation
T
Temperature
TSP
Total Suspended Particles
Hm
Micrometer
UV
Ultraviolet
V
Volt
VOC
Volatile Organic Compound
-------
Executive Summary
Background
The Office of Research and Development (ORD) at EPA has been conducting a variety of both laboratory
and field-based evaluations/deployments of air quality sensors and other Next Generation Air Monitors
(NGAM). These efforts have included laboratory and/or field evaluations of numerous NGAM devices for
measuring nitrogen dioxide (N02), ozone (03), particulate matter (PM), volatile organic compounds (VOC),
and sulfur dioxide (S02) (Williams 2015a, Williams 2015b, Williams 2014a, Williams 2014b). The ORD has
published findings from some of these efforts through its Air Sensor Toolbox for Citizen Scientists,
Researchers and Developers website (www.epa.gov/air-sensor-toolbox). During recent years, air pollution
sensor technology has advanced significantly. Specifically, sensors that currently are being developed are
much smaller, more lightweight, and lower in cost than traditional ambient air monitoring systems. The
advent of these types of sensors present an opportunity to advance EPA's strategic goals, including those of
community monitoring and environmental justice. One of the potential benefits of this type of technology is
the ability to deploy a larger number of sensors across a small geographic area (e.g., a neighborhood) and
collect data with a level of spatial and temporal resolution that would be cost-prohibitive using traditional
monitoring methods. Prime examples of such efforts include the Citizen Science Air Monitor (CSAM; Barzyk
2016, Williams 2015c) and the AirMapper (U.S. EPA, 2016) devices. All the aforementioned examples
involved ORD collaboration with regional offices and local communities and/or state air quality agencies. To
the greatest extent possible, technology insights from such recent projects were leveraged in this study to
reduce sensor pod development costs as well as provide for project timeline savings. This project supports
the development of low-cost sensor technologies that can be used for community monitoring.
Study Objectives
The main objective of this project was to fully characterize the performance of the low-cost air quality
sensors devices developed by the EPA (herein referred to as pods, sensor pods, or CSAM) and to better
understand their potential for community monitoring applications. The performance of the sensor pods
developed by the EPA was evaluated and characterized in the field (i.e., at one of SCAQMD's air monitoring
stations) and in a controlled laboratory environment (i.e., AQ-SPEC's characterization chamber). After
evaluating the sensors' performance, the pods were dispersed in a nodal pattern to study the spatial and
temporal patterns of multiple criteria pollutants in Southern California.
Study Approach
The EPA developed and assembled ten identical pods containing low-cost commercial-off-the-shelf (COTS)
sensors for measuring PM, 03, relative humidity (RH), and temperature (T). Each pod has AC and/or solar
power functionality and internal data storage. The basic operational status of the pods was established at
EPA's Research Triangle Park (RTP) laboratory in North Carolina. A research operating p rotocol (ROP) for
operating the sensor pod was developed by EPA staff. Once the pods were deemed operational and fully
functional, they were shipped to the AQ-SPEC group, which conducted an intensive field and laboratory
performance evaluation of the PM2.5, 03, RH, and T sensors against the corresponding reference monitors.
Ultimately, the pods were placed in a nodal pattern in areas of high interest to EPA Region 9 (i.e.,
adjacent to freeways, existing regulatory monitoring stations, and associated neighborhoods including
schools, as well as emissions sources such as marine ports, airports, etc.) and operated for approximately
three months from January 2017 to April 2017. The components of the EPA-designed sensor pods, selected
for their known capabilities of being integrated into a sensor pod arrangement, are defined in Table A.l.
xiv
-------
Table A.1. Sensor components in each CSAM Sensor Pod developed by EPA
Sensor / Manufacturer
Parameters Measured
Approximate total cost (per
pollutant for multi-pollutant
OPC-N2 (AlohaSense)
PMi, PM2.5and PMio
$500
SM-50 (Aeroqual)
03(Ozone)
$500
Adafruit AM 2315
Relative humidity (RH)
$200
Adafruit AM 2315
Temperature
$200
Grape Solar 1289
Solar panel
$500
Arduino Mega with Adafruit SD
Microprocessor
$400
EPA has prior experience assembling integrated sensor devices (e.g., CSAM, AirMapper) and sharing their
features with all interested parties. Figures A.l and A.2 below depict the basic CSAM pod and deployment
components (sensor box, tripod, solar panel) that were used for the project.
Figure A.1. Sensor pod assembly Figure A.2. Fully assembled pod with solar
panel, battery cell, and tripod.
Each sensor component was operated according to the manufacturer's recommendations. On-board data
storage (secure digital [SD] card) was used to minimize data collection issues. A weatherproof enclosure was
used to contain all sensor components, with inlets protruding from the base of the enclosure. The placement
of the solar panel was designed to act as a shading device to reduce the internal operating temperature of
the pod. A stable tripod base provided the means to secure all components and allowed for consistent
deployment across the various geographical locations. Once the devices were determined to be operating
in an acceptable manner, they were delivered to AQ-SPEC for laboratory and field evaluation and then for
subsequent field deployment.
The AQ-SPEC team was responsible for evaluating the sensor pods developed by ORD and then operating
the pods in a nodal (spatial) pattern across Southern California in specific areas of interest to EPA Region 9
and SCAQMD. The evaluation of the sensor pods was conducted at SCAQMD field and laboratory locations.
The individual sensor components selected for the pods had been evaluated previously as part of the
CAIRSENSE field tests (Jiao 2016, Jiao 2015) under ambient conditions. The performance of these same
sensors in California (field/laboratory) was expected to be similar to these earlier findings. Even though both
xv
-------
the 0PC-N2 and the Aeroqual SM-50 sensors had been tested previously by the EPA over several months, a
full (e.g., field and laboratory) evaluation of these sensors at this geographic scope had not been performed
prior to this effort. The current project described herein adds to EPA's knowledge base of these sensors and
their performance characteristics.
This project consisted of both laboratory and field NGAM evaluation scenarios over the lifespan of the
research project. The PM and ozone gas pollutant sensors were compared with the appropriate Federal
Equivalent Method or Federal Reference Method (FEM or FRM, respectively) air pollutant measurement
systems. Field and laboratory evaluations involving FEM/FRM monitors operated by SCAQMD's personnel
were conducted under AQ-SPEC's internal quality assurance guidelines.
Conclusions
The ten CSAM pods, each one equipped with i) a PM2.5 sensor, model OPC-N2, by Alphasense, UK; ii) an
ozone sensor, model SM-50, by Aeroqual, New Zealand, and; iii) a temperature/relative humidity sensor,
model AM2315, by Adafruit, were co-located and field tested alongside reference instruments for two
months at one of SCAQMD's fixed ambient monitoring stations in Riverside-Rubidoux (RIVR), California. One
CSAM pod developed operational issues at the onset and was not deployed. The other nine CSAM pods
proved to be very reliable, with excellent data recovery and overall showed only a modest intra-model
variability for PM2.5 and a low intra-model variability for ozone, temperature, and relative humidity. Three
CSAM Pods carried PM sensors that indicated a potential mischaracterization of ambient PM2.5 mass
concentration measurements compared to those from the reference instruments. The CSAM PM2.5 sensor
data for seven of the nine units tested showed a modest correlation (R2: 0.40-0.60) with substantially more
costly FEM instruments (i.e., GRIMM and BAM), and overestimated the 5-minute average PM2.5
measurements by approximately 75% as measured by the FEM GRIMM. The PM2.5 data from two units
showed a poor correlation with the FEM due to a potential PM sensor malfunction with the sensor's optics
during the co-location testing period. The CSAM ozone sensor data for all nine pods showed excellent
correlation (R2: 0.80 - 0.97) with the corresponding measurements of an FRM instrument (i.e., Thermo 49i)
but underestimated the 5-minute average ozone concentrations by 10 to 50% as measured by the FRM
instrument. The CSAM temperature and relative humidity measurements from the nine sensors correlated
very well with the corresponding RIVR data (R2 > 0.80 and R2 > 0.91 for temperature and relative humidity,
respectively). Upon completion of the field co-location testing, two of the CSAM pods were subsequently
brought back to the AQ-SPEC laboratory for further testing. Under controlled environmental conditions in
the laboratory chamber, the two CSAM Pods were very reliable and showed excellent correlation with the
FEM GRIMM and FRM ozone instruments (R2 > 0.99 and R2 > 0.95 for PM2.5 and ozone, respectively).
However, the two PM sensors in the CSAM pods consistently overestimated the FEM readings for a wide
range of PM2.5 mass concentrations. On the contrary, the two ozone sensors in the two CSAM Pods
underestimated the FRM readings for a wide range of ozone concentrations at average ambient weather
conditions. Both the PM and ozone sensors in the two CSAM Pods showed good precision under more
"extreme" weather conditions. When challenged with a range of nitrogen dioxide (N02) concentrations, the
two ozone sensors in the two CSAM Pods were not affected by the presence of a potential ozone interferent
gas, reporting zero values, which indicates that the two sensors may not respond to such interferences in
the ambient air.
The CSAM Pods then were deployed successfully at nine locations across southern California in three
distinct areas including Long Beach, Jurupa Valley, and Coachella Valley. The extended deployment and
subsequent data analysis revealed some of the concerns pertaining to the long-term sensor deployment of
low-cost sensors. The AQ-SPEC field and laboratory evaluation process provides a short-term study to
xvi
-------
identify sensor performance over a specific two-month period with specific ambient conditions. As
ambient air quality studies begin to involve low-cost air quality sensors for extended time periods,
identifying and quantifying sensor response degradation and characterizing performance over time
becomes crucially important to provide meaningful data and results. The reliability of the sensor is also
important, as reductions in data recovery and sensor performance over time significantly increase the
complexity related to comparing spatial and temporal differences between sensor locations. Throughout
this study, the OPC-N2 was found to have significant data losses with low data recovery at several sites.
The ozone sensor had high data recovery, but its readings were found to degrade over the deployment
period (three months in this case) when compared with reference instrumentation. Validating these
sensors' performance and keeping them calibrated over time will be a significant challenge for agencies
and organizations attempting to deploy sensor networks to monitor ambient air quality over a wide area.
The ability to access sensor data and use real-time online dashboards remotely would provide the
opportunity to automate validating procedures for cloud-based applications to properly characterize sensor
failures and quantify data loss.
Lessons learned during the study period suggest that low-cost sensors like those employed here may have a
role in air quality monitoring under highly controlled situations. Even so, the fact that a sensor's
responsiveness can change over a fairly brief time or under varying meteorological conditions suggests that
a sophisticated quality assurance program for their deployment would be useful. We observed significant
changes in gas phase sensor response (~ 6 months of age) although most of the sensors' manufacturers
suggest lifespans of approximately 1 year. Our experiences during this study confirmed our previous
experience with these types of sensors relative to their actual lifespans.
-------
Introduction
The AQ-SPEC group evaluated the field and laboratory performance of ten Citizen Science Air Monitor
(CSAM) units (sensor pods #401-410) developed by the EPA to measure ambient PM2.5, ozone, temperature
and relative humidity. The ten CSAM units were tested in a co-location setting at the SCAQMD Rubidoux
(RIVR) Air Monitoring Site (AMS) between 11/9/2016 and 01/05/2017. Results of the co-location
performance testing are presented in Section 1. The performance of two of the CSAM units (406 and
409) were evaluated in post co-location testing in the AQ-SPEC laboratory chamber. The results of this
evaluation are summarized in Section 2. The CSAM pods were deployed throughout Southern California
for approximately three months in three distinct geographical areas including Long Beach, Jurupa Valley,
and Coachella Valley. The results of this field deployment are summarized in Section 3 of this report.
1. Co-location Testing
Through the AQ-SPEC program, sensors are typically tested in the field "out-of-the-box", without prior
modifications, calibrations, or checks, including zero, span, or precision checks. The sensors were
operated according to the manufacturer's user guide or manual. If specified by the sensor
manufacturer's user guide or manual, the sensors may have undergone routine maintenance
throughout the study. Such routine maintenance may have included but was not limited to filter
replacement, zero calibration, flow rate checks, date/time synchronization, and battery change. The air
monitoring stations where the field testing was conducted was equipped with FRM, FEM, or best
available technology (BAT) air monitoring equipment that was routinely used to measure the
ambient concentrations of gaseous or particle pollutants for regulatory purposes. The low-cost sensors
were deployed side by side with the FRM, FEM, or BAT monitoring equipment. The data obtained from the
two monitoring methods are compared and used as the primary tool in the AQ-SPEC evaluation process of
low-cost sensors. To ensure statistically relevant data sets, the sensors were deployed in triplicate and for
a period of two months. Deployments in triplicate allow for a statistically robust intra-model comparability
between the three sensors and for the ability to detect potential sensor failures or malfunctions. Sensors
that are ruggedized and designed for ambient air monitoring purposes are typically mounted outside on
the protective railing of the AMS. Sensors that are not ruggedized for inclement weather and are designed
for ambient air monitoring conditions are deployed in a custom -built sensor shelter.
1.1. Methods
Before a sensor's field evaluation begins, a bench test at SCAQMD is performed. The bench test involves:
• Reviewing the sensor documentation, including the manual or operating procedures
• Evaluating the power options: cable, battery, and/or solar
• Evaluating the data acquisition options: device internal storage, laptop data logging, cloud based
• Evaluating the data output format to ensure a usable format
• Evaluating the functionality of the On/Off switch to test whether sensors turn on properly
Upon successful completion of the bench test, the sensors were brought to the SCAQMD's Riverside-
Rubidoux (RIVR), CA AMS. Since the CSAM pods were ruggedized for inclement weather, the CSAM pods
were set up outside of the AMS, as shown in Figure 1.1. CSAM pods #401 through #407 were mounted on
tripods and were configured with solar power components. CSAM pods #408- through #410 were
configured for 120V power and mounted on a safety railing next to the other CSAM units at the AMS
1
-------
(Figure 1.2). It should be stated that all CSAMs had the capability of operating on land-based power.
Although some of the pods had solar capability, all the pods were eventually operated on land
power for consistency. The CSAM's inlets and reference monitors were within ± 2 meters of the
same height. The CSAM pods were then exposed to ambient air for more than two months from
10/13/2016 to 01/05/2017 (further details on the AQ-SPEC's Field Setup and Testing Protocol can be found
in Appendix B).
Figure 1.1. Co-location set-up for CSAM units #401 - #407 at Riverside-Rubidoux AMS
Figure 1.2. Co-location set-up for CSAM units #408 - #410 at Riverside-Rubidoux AMS
To ensure that the sensors were not crowding one another, care was taken to place them a minimum of
8 inches apart. The sensors were shielded from rain using their own weather-protected NEMA enclosures.
After deployment, the AQ-SPEC personnel checked each of the sensors on a biweekly basis to ensure proper
functionality and that the data were downloaded periodically from the sensor's SD card. All data collected
by the sensors were compared to those collected from the reference monitors at that site.
The AQ-SPEC reference instrumentation at the Riverside-Rubidoux fixed AMS used for this study included:
A GRIMM Dust Monitor (model EDM 180, Ainring, Germany): The EDM 180 spectrometer provides high-
resolution, real-time aerodynamic measurements of PMio, PM2.s, PMi.o, total suspended particulate (TSP),
and PM coarse (PMC). The EDM 180 measures light-scattering at a resolution time of one minute, costs more
2
-------
than $25,000 and is designated as class III equivalent method EQPM-0311-195 by the U.S. EPA.
A Met One Instruments Particulate Monitor (model BAM-1020 PMio & PM2.5, Grants Pass, OR). The BAM-
1020 automatically measures and records airborne particulate concentration levels using the principle of
beta ray attenuation. This method provides a simple determination of concentration in units of milligrams
or micrograms of particulate per cubic meter of air, at a time resolution of one -hour, and costs more than
$20,000. The BAM-1020 is designated as an equivalent method for PM10 EQPM-0798-122 and an equivalent
method for PM2 5 EQPM-0308-170 by the U.S. EPA.
A Thermo Fischer Scientific UV Photometric O3 Analyzer (Model 49/', Franklin, MA): The Model 49/ operates
on the principle that ozone (O3) molecules absorb UV light at a wavelength of 254 nm, has a time
resolution of one minute, and costs more than $7,000. The Model 49/ is designated as an equivalent
method for the measurement of ambient concentrations of ozone EQOA-0880-047 by the U.S. EPA.
A Rotronic AG HygroClip2-S3 Temperature/RH probe (Hauppage, NY) was used to measure the ambient
temperature and relative humidity at the SCAQMD fixed ambient air monitoring Riverside-Rubidoux station
(see appendix B for manual reference).
Collocation testing was based on a side-by-side comparison between the ten sensor pods and the FRM/FEM
instruments that were measuring the same pollutant(s). A series of performance-related parameters that
would affect air quality measurements in the field were evaluated. These parameters included:
• Intra-model variability
• Data recovery
• Linear correlation coefficient (R2)
A detailed description of the methodology for estimating each evaluation parameter as well as detailed
experimental procedures for sensor testing are described in sections 3.1 and 3.2 of the AQ-SPEC Field Setup
and Testing Protocol (see appendix B).
1.2. Results
The ten EPA CSAM units were co-located at the RIVR AMS and operated alongside EPA-approved FEM and FRM
instruments from October 13,2016 to January 5,2017. Due to problems with solar power generation, consistent
data recovery for the CSAM units was not obtained until November 9, 2016. As a result, the data obtained for
the comparison of the temperature, relative humidity, and ozone levels extended from November 9, 2016 to
January 5, 2017. Additionally, because the Grimm 180 EDM was undergoing calibration by Grimm Technologies
during the beginning of the study, the particulate matter data comparison period was delayed until December
8, 2016 and extended to January 5, 2017. CSAM units #401, #402, #403, #404, #406, #407, #408, #409 and #410
were included in the co-location analysis. CSAM unit #405 was removed from the analysis due to an error with
the time stamp readings. The unit was sent back to EPA for re-configuration and was subsequently included in
the field deployment portion of the study.
1.2.1. Ambient pm2.5
CSAM PM data validation and recovery
Standard QA/QC procedures were used to validate the data collected from the PM2.5 sensors (model OPC-
N2, AlphaSense, UK) in the CSAM units. Obvious outliers, negative values, and invalid data-points
were eliminated from the data-set. Data recovery for the PM2.5 sensor from all nine CSAM units was >
99.1%. Descriptive and correlation statistics for the nine units are presented in Tables l.la-b below.
3
-------
Table 1.1a. Descriptive statistics for PM2.5 sensors in the CSAMs and the FEM GRIMM reference
instrument (5-minute average)
PM2.5 (Hg/m3)
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
GRIMM
mean
3.2
41.0
21.2
4.4
20.3
21.7
23.5
20.5
30.8
15.1
median
3.3
17.8
5.7
0.0
5.6
5.7
5.7
5.1
9.1
9.6
STDEV
0.8
52.2
35.4
12.8
33.6
37.0
40.8
36.1
48.5
13.4
Count (#)
7947
7948
7948
7946
7947
7946
7947
7947
7947
7967
Recovery {%)
99.7
99.8
99.8
99.7
99.7
99.7
99.7
99.7
99.7
100
Table 1.1b. Correlation statistics for PM2.5 sensors in the CSAMs against the FEM GRIMM reference
instrument (5-minute average), [FEM= (slope*sensor reading) + intercept]
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
Slope
0.4849
0.1903
0.2715
0.2892
0.2889
0.2573
0.2237
0.2504
0.1974
Intercept
13.595
7.3293
9.3622
8.8361
9.2564
9.5382
9.8765
10.006
9.0523
R2
0.0008
0.5488
0.5138
0.2356
0.5247
0.5039
0.4644
0.4551
0.5098
CSAM PM intra-model variability
Moderate measurement variations were observed between the nine CSAM units for PM2.5 mass
concentrations (ng/m3). Figure 1.3 shows a modest intra-model variability (mean and median values)
between seven of the CSAM PM25 units, excluding units #401 and #404.
PM25
¦ mean + SD ¦ median
m£ 100
Figure 1.3. Intra-model variability for nine of the ten CSAM PM2.5 sensors tested. Vertical bars represent
the standard deviation for each mean value
FEM data validation and recovery
Standard QA/QC procedures were used to validate the FEM data collected (i.e., obvious outliers, negative
values, and invalid data-points were eliminated from the data-set). PM2.5data recovery was 100% for the
GRIMM and 96 % for the BAM reference instruments.
Excellent correlation was observed between the 1-hour average mass concentrations of the two
4
-------
equivalent methods for PM2.s (R2: 0.945). The two FEM instruments tracked well with each other
throughout the entire co-location test period (see Figures 1.4 and 1.5).
PM2 5 (|ig/m3; 1-hr mean)
80
^ 60
<
- 40
£ 20
0
0 20 40 60 80
FEM GRIMM
Figure 1.4. Correlation coefficient (R2) plot for the 1-hour average PM2.5 measurements by the FEM
GRIMM and FEM BAM units (1-hour average)
FEM GRIMM vs FEM BAM (PM2 5; 1-hr mean)
— FEM GRIMM FEM BAM
80
Figure 1.5. Time-series plot of PM2.5 measurements from the FEM GRIMM vs FEM BAM units (1-hour
average)
CSAM vs FEM GRIMM fPM2.s moss; 5-min mean)
Most of the 5-minute average CSAM PM2.5 mass measurements showed a modest correlation with the
corresponding FEM GRIMM data (R2 > 0.45) (Table 1.2.). The PM2.5 from units #401 and #404 did not
correlate well with the FEM GRIMM data. The nine sensor units did not appear to track well the diurnal
PM variations recorded by the FEM GRIMM instrument (Figures 1.6-1.7).
5
-------
CSAM PM vs FEM GRIMM (PM2 5)
FEM GRIMM Unit 401 Unit 402 Unit 403 Unit 404
400
Figure 1.6. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM
instrument (5-minute average)
CSAM PM vs FEM GRIMM (PMZ s)
—FEM GRIMM —Unit406 —Unit 407 —Unit408 —Unit409 —Unit410
400
Figure 1.7. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM
instrument (5-minute average)
Table 1.2. Correlation Coefficient (R2) matrix for the 5-minute average PM2.5mass concentrations
measured by the FEM and CSAM units.
2
R
FEM
GRIMM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
GRIMM
1
Unit 401
0.0008
1
Unit 402
0.5488
0.0002
1
Unit 403
0.5138
0.0007
0.9831
1
Unit 404
0.2356
0.0676
0.9655
0.9815
1
6
-------
2
R
FEM
GRIMM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
Unit 406
0.5247
0.0010
0.9775
0.9945
0.9827
1
Unit 407
0.5039
0.0001
0.9820
0.9931
0.9806
0.994
1
Unit 408
0.4644
0.0004
0.9736
0.9708
0.9583
0.961
0.972
1
Unit 409
0.4551
0.0001
0.9739
0.9783
0.9587
0.971
0.979
0.989
1
Unit 410
0.5098
0.0001
0.9857
0.9762
0.9669
0.968
0.976
0.991
0.989
1
CSAM vs FEM GRIMM (PM2.s moss; 1-hour mean)
Most of the 1-hour average CSAM PM2.5 mass measurements showed a modest correlation with the
corresponding FEM GRIMM data (R2 > 0.46) (Table 1.3). The PM2.5 measurements from units #401 and #404
did not correlate with the FEM GRIMM data. The nine CSAM units did not appear to track well the diurnal
PM variations recorded by the FEM GRIMM instrument (Figures 1.8-1.9).
CSAM PM vs FEM GRIMM (PM25)
350
300
E
250
3
c
o
% 200
c
V
j 150
c
? 100
jc
i-i
50
0
12/7/16 12/12/16 12/17/16 12/22/16 12/27/16 1/1/17 1/6/17
Figure 1.8. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM
instrument (1 -hour average)
CSAM PM vs FEM GRIMM (PMj 5)
—FEM GRIMM —Unit 406 —Unit 407 Unit 408 —Unit 409 —Unit 410
350
Figure 1.9. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM
instrument (1-hour average)
7
—FEM GRIMM —Unit 401 —Unit 402 Unit 403 —Unit 404
-------
Table 1.3. Correlation Coefficient (R2) matrix for the 1-hour average PM2.5 mass concentrations
measured by the FEM and CSAM units
R2
FEM
GRIMM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
GRIMM
1
Unit 401
0.0006
1
Unit 402
0.5641
0.0001
1
Unit 403
0.5269
0.0009
0.9885
1
Unit 404
0.2541
0.0715
0.9880
0.9977
1
Unit 406
0.5376
0.0011
0.9846
0.9983
0.9971
1
Unit 407
0.5164
0.0001
0.9874
0.9972
0.9978
0.997
1
Unit 408
0.4762
0.0003
0.9823
0.9783
0.9832
0.969
0.980
1
Unit 409
0.4670
0.0001
0.9813
0.9855
0.9831
0.978
0.985
0.992
1
Unit 410
0.5234
0.0001
0.9921
0.9829
0.9926
0.976
0.981
0.994
0.990
1
CSAM vs FEM GRIMM (PM2.5 mass; 24-hour mean)
Most of the 24-hour average CSAM PM2.5mass measurements showed a modest correlation with the
corresponding FEM GRIMM data (R2 > 0.49) (Table 1.4). The PM2.5 measurements from units #401 and #404
did not correlate with the FEM GRIMM data. The nine CSAM units did not appear to track well the diurnal
PM variations recorded by the FEM GRIMM instrument and tended to overestimate the FEM measurements
(Figures 1.10-1.11).
CSAM PM vs FEM GRIMM (PM25)
— FEM GRIMM —Unit 401 —Unit 402 —Unit 403 —Unit 404
160
Figure 1.10. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM
instrument (24-hour average)
8
-------
CSAM PM vs FEM GRIMM (PM2 5)
— FEM GRIMM —Unit 406 —Unit 407 —Unit 408 —Unit 409 —Unit 410
Figure 1.11. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM instrument
(24-hour average)
Table 1.4. Correlation Coefficient (R2) matrix for the 24-hour average PM2.5 mass concentrations
measured by the FEM and CSAM units
2
R
FEM
GRIMM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
GRIMM
1
Unit 401
0.0664
1
Unit 402
0.5946
0.0365
1
Unit 403
0.5782
0.0539
0.9900
1
Unit 404
0.3807
0.1077
0.9947
0.9980
1
Unit 406
0.5985
0.0639
0.9817
0.9976
0.9982
1
Unit 407
0.5698
0.0522
0.9888
0.9994
0.9985
0.997
1
Unit 408
0.4933
0.0214
0.9836
0.9695
0.9961
0.951
0.969
1
Unit 409
0.5000
0.0287
0.9885
0.9824
0.9960
0.969
0.984
0.995
1
Unit 410
0.5415
0.0257
0.9935
0.9790
0.9975
0.964
0.978
0.996
0.994
1
CSAM vs FEM BAM (PM2.5 mass; 1-hour mean)
Most of the 1-hour average CSAM PM2.5 mass measurements showed a modest correlation with the
corresponding FEM BAM data (R2 > 0.40) (Table 1.5). The PM2.5 measurements from units #401 and #404
did not correlate with the FEM BAM data. The nine sensor units did not appear to track well the diurnal
PM variations recorded by the FEM BAM instrument and tended to overestimate the FEM
measurements (Figures 1.12-1.13).
9
-------
400
CSAM PM vs FEM BAM (PM2 5)
— FEM —Unit 401 —Unit402 —Unlt403 —Unit404
1 300
u
3
I 250
-------
Table 1.5. Correlation Coefficient (R2) matrix for the 1-hour average PM2.5 mass concentrations
measured by the FEM and CSAM units
R2
FEM
BAM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
BAM
1
Unit 401
0.0012
1
Unit 402
0.4941
0.0001
1
Unit 403
0.4612
0.0009
0.9885
1
Unit 404
0.1339
0.0715
0.9880
0.9977
1
Unit 406
0.4730
0.0011
0.9846
0.9983
0.9971
1
Unit 407
0.4522
0.0001
0.9874
0.9972
0.9978
0.9970
1
Unit 408
0.4058
0.0003
0.9823
0.9783
0.9832
0.9687
0.9795
1
Unit 409
0.4027
0.0001
0.9813
0.9855
0.9831
0.9780
0.9853
0.9920
1
Unit 410
0.4499
0.0001
0.9921
0.9829
0.9926
0.9756
0.9814
0.9942
0.9899
1
CSAM vs FEM BAM (PM2.s mass; 24-hour mean)
Most of the 24-hour average CSAM PM2.5mass measurements showed a modest correlation with the
corresponding FEM BAM data (R2 > 0.44) (Table 1.6). The PM2.5 measurements from units #401 and #404
did not correlate with the FEM BAM data. The nine sensor units did not appear to track well with the
diurnal PM variations recorded by the FEM BAM instrument and tended to overestimate the FEM
measurements (Figures 1.14-1.15).
IS"
M
% 120
1
I 10
I 8
I
i 6
I
I «
CSAM PI
-Unit 401
2.5J
KB -
1;,,•>1;:1 u i.\ I*, io u<"> ,•< 1."4[>>. \ i-'i, 1, l-'i1
Figure 1.14. Time-series plot of PM2.5 measurements from units #401 through #404 and the FEM
instrument (24-hour average)
11
-------
CSAM PM vs FEM BAM (PM2 5)
— FEM —Unit 406 —Unit 407 —Unit 408 —Unit 409
— Unit 410
160
Figure 1.15. Time-series plot of PM2.5 measurements from units #406 through #410 and the FEM
instrument (1-hour average)
Table 1.6. Correlation Coefficient (R2) matrix for the 1-hour average PM2.5 mass concentrations
measured by the FEM and CSAM units
R2
FEM
BAM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM BAM
1
Unit 401
0.0546
1
Unit 402
0.5451
0.0365
1
Unit 403
0.5377
0.0539
0.9900
1
Unit 404
0.1676
0.1077
0.9947
0.9980
1
Unit 406
0.5619
0.0639
0.9817
0.9976
0.9982
1
Unit 407
0.5307
0.0522
0.9888
0.9994
0.9985
0.9972
1
Unit 408
0.4407
0.0214
0.9836
0.9695
0.9961
0.9510
0.9692
1
Unit 409
0.4535
0.0287
0.9885
0.9824
0.9960
0.9687
0.9838
0.9954
1
Unit 410
0.4881
0.0257
0.9935
0.9790
0.9975
0.9643
0.9777
0.9963
0.9945
1
Pod #405 was removed due to frequent erroneous time stamp readings and returned to the EPA for repair and
re-configuration. The initial period of the co-location (October 13, 2016 to November 9, 2016) was littered
with errors due to power issues with units #401 through #407. These seven units, which were configured
with a solar panel and battery for power generation, did not operate properly. The 3-amp fuse located
between the charge controller and the battery was not sufficient to sustain the loads experienced. As a
result, fuses failed frequently, which subsequently led to dead batteries and loss of power. The low power
situations led to real-time clock errors and an overall data loss. Power adaptors and cords were purchased
to remedy the solar power issues, and CSAM units #401 through 407 were configured to 120V power on
November 9, 2017. Data recovery improved once 120V power supplies were used for all nine of the CSAM
units. As a result, November 9, 2017 was considered the "official" start date for the field co-location study.
Another error that occurred for unknown reasons was that the CSAM units would stop logging data and the
display would indicate VOC rather than ozone. Due to the Grimm 180 EDM undergoing calibration at the
12
-------
factory, the particulate matter data collocation period for PM was set from December 8, 2016 to
January 5, 2017 for data analysis to match the time period that the Grimm data were available.
Overall, the CSAM PM25data were very reliable, with data recovery close to 100% for all units tested, and
were characterized by modest intra-model variability. The CSAM PM2.5 sensor data for seven of the units
(#402, #403, #406, #407, #408, #409, and #410) showed a modest correlation (R2: 0.40 - 0.60) with the
corresponding measurements collected using substantially more expensive FEM instruments (i.e., GRIMM,
BAM) at 5-minute, 1-hour, and 24-hour time resolutions. The CSAM PM2.5 sensors for pods #401 and #404
did not correlate well with the FEM instruments at any time resolution. Overall, the CSAM PM2.5 sensors
overestimated the mass concentration measurements from the FEM GRIMM and FEM BAM reference
instruments.
1.2.2. Ambient Ozone
Data validation and recovery
Standard QA/QC procedures were used to validate the collected data from the ozone sensors (model SM-50,
Aeroqual, New Zealand) in the CSAM units. Obvious outliers, negative values, and invalid data-points were
eliminated from the data-set. The data recovery for ozone from eight of the CSAM units was 99.3%, for
unit #402 was 77.5%, and for the FEM reference instrument was 93.3%, Descriptive and correlation
statistics for the nine units and the FRM instrument are presented in Tables 1.7a-b below.
Table 1.7a. Descriptive Statistics for ozone from the CSAMs and the FEM instrument
Ozone (ppb)
FEM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
mean
18.9
17.3
15.5
19.6
12.5
18.6
20.8
23.3
12.3
17.2
median
16.9
14.3
14.9
17.6
10.9
15.4
19.2
21.6
9.9
14.9
SD
15.9
14.6
6.9
9.6
9.2
13.5
8.7
10.1
11.6
11.3
Count (#)
15319
16304
12733
16307
16304
16305
16304
16303
16304
16304
Recovery {%)
93.3
99.3
77.5
99.3
99.3
99.3
99.3
99.3
99.3
99.3
Table 1.7b. Correlation statistics for ozone from the CSAMS against the FEM instrument (5- minute
average) [FEM = (slope*sensor reading) + intercept]
#401
#402
#403
#404
#406
#407
#408
#409
#410
Slope
1.0621
2.0249
1.594
1.5779
1.0978
1.7731
1.5367
1.2618
1.3563
Intercept
0.0844
-13.256
-12.845
-1.3013
-1.9282
-18.511
-17.457
3.0019
-4.8449
R2
0.9524
0.8563
0.9394
0.8367
0.877
0.9379
0.9538
0.8549
0.9339
CSAM Ozone; intra-model variability
Low measurement variations were observed between the nine CSAM units for ambient ozone
concentrations (ppb). Figure 1.16 shows the intra-model variability (mean and median values) in the nine
CSAM ozone sensors.
13
-------
Si
Q.
Q.
o 40
| 30
g 20
o
^ 10
ra
5 0
c
£
Ozone
mean ± STDEV
median
I
1/1 cr ^ ^ ^ ^ ^ ^
Figure 1.16. Intra-model variability in nine CSAM ozone sensors
CSAM vs FEM (Ozone; 5-minute mean)
AN 5-minute average CSAM ozone measurements from the nine sensors correlated very well with the
corresponding FEM data (R2 > 0,83) (Table 1.8). The nine CSAM units appeared to track well with the diurnal
ozone variations recorded by the FEM instrument but underestimated the 5-minute average ozone
concentrations by 10 to 50% as measured by the FEM (Figures 1.17-1.18).
CSAM vs FEM (Ozone)
— FEM —Unit 401 —Unit 402 —Unit 403 —Unit 404
70
HI
11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.17. Time-series plot of ozone measurements from units #401 through #404 and the FEM
instrument (5-minute average)
14
-------
CSAM vs FEM (Ozone)
— FEM —Unit 406 —Unit 407 Unit 408 —Unit 409 —Unit 410
70
60
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.18. Time-series plot of ozone measurements from units #406 through #410 and the FEM
instrument (5-minute average)
Table 1.8. Correlation Coefficient (R2) matrix for the 5-minute average ozone concentrations
measured by the FEM and CSAM units
2
R
FEM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
1
Unit 401
0.9524
1
Unit 402
0.8563
0.9219
1
Unit 403
0,9,394
0.9780
0.9309
1
Unit 404
0.8367
0.9091
0.9499
0.8992
1
Unit 406
0.8770
0.9311
0.9141
0.9468
0.8510
1
Unit 407
0.9379
0.9796
0.9308
0.9851
0.8920
0.9698
1
Unit 408
0.9538
0.9881
0.9109
0.9831
0.8830
0.9439
0.9881
1
Unit 409
0.8549
0.9125
0.9575
0.9314
0.8816
0.9728
0.9520
0.9174
1
Unit 410
0.9339
0.9860
0.9458
0.9822
0.9291
0.9552
0.9874
0.9819
0.9500
1
CSAM vs FEM (Ozone; 1-hour mean)
All 1-hour average CSAM ozone measurements from the nine sensors correlated very well with the
corresponding FEM data (R2 > 0.86) (Table 1,9). The nine sensor units appeared to track well with the diurnal
ozone variations recorded by the FEM instrument (Figures 1.19-1.20).
15
-------
70
CSAM vs FEM (Ozone)
—FEM —Unit 401 —Unit402 —Unit403 —Unit404
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.19. Time-series plot of ozone measurements from units #401 through #404 and the FEM
instrument (1-hour average)
CSAM vs FEM (Ozone)
—FEM —Unit 405 —Unit 406 —Unit 407 —Unit 408 —Unit 409 —Unit 410
70
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.20. Time-series plot of ozone measurements from units #406 through #410 and the FEM
instrument (1-hour average)
16
-------
Table 1.9. Correlation coefficient (R2) matrix for the 1-hour average ozone concentrations measured
by the FEM and CSAM units
2
R
FEM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
1
Unit 401
0.9683
1
Unit 402
0.8891
0.9288
1
Unit 403
0.9653
0.9825
0.9355
1
Unit 404
0.8659
0.9121
0.9544
0.9013
1
Unit 406
0.8999
0.9327
0.9189
0.9496
0.8517
1
Unit 407
0.9622
0.9830
0.9353
0.9885
0.8936
0.9722
1
Unit 408
0.9760
0.9905
0.9170
0.9874
0.8853
0.9457
0.9913
1
Unit 409
0.8796
0.9147
0.9618
0.9336
0.8821
0.9748
0.9538
0.9192
1
Unit 410
0.9590
0.9894
0.9498
0.9853
0.9309
0.9567
0.9895
0.9846
0.9511
1
CSAM vs FEM (Ozone; 8-hour mean)
All 8-hour average CSAM ozone measurements from the nine sensors correlated very well with the
corresponding FEM data (R2 > 0.85) (Table 1.10). The nine sensor units appeared to track well with the
diurnal ozone variations recorded by the FEM instrument (Figures 1.21-1.22).
CSAM vs FEM (Ozone)
—FEM —Unit 401 —Unit 402 Unit 403 —Unit 404
50
Figure 1.21. Time-series plot of ozone measurements from units #401 through #404 and the FEM
instrument (8-hour average)
17
-------
CSAM vs FEM (Ozone)
—FEM —Unit 406 —Unit 407 —Unit 408 —Unit 409 —Unit 410
70
60
-Q
g: so
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.22. Time-series plot of ozone measurements from units #406 through #410 and the FEM
instrument (8-hour average)
Table 1.10. Correlation coefficient (R2) matrix for the 8-hour average ozone concentrations measured
by the FEM and CSAM units
2
R
FEM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
1
Unit 401
0.9742
1
Unit 402
0.8883
0.9229
1
Unit 403
0.9684
0.9813
0.9285
1
Unit 404
0.8556
0.8998
0.9594
0.8878
1
Unit 406
0.8852
0.9139
0.8901
0.9383
0.8194
1
Unit 407
0.9617
0.9778
0.9237
0.9884
0.8761
0.9666
1
Unit 408
0.9775
0.9874
0.9097
0.9879
0.8707
0.9332
0.9897
1
Unit 409
0.8593
0.8912
0.9376
0.9192
0.8553
0.9698
0.9441
0.8997
1
Unit 410
0.9603
0.9859
0.9477
0.9856
0.9259
0.9443
0.9869
0.9819
0.9367
1
CSAM vs FEM (Ozone; 24-hour mean)
All 24-hour average ozone measurements from the nine CSAMs correlated very well with the corresponding
FEM data (R2 > 0.80) (Table 1.11). The nine CSAMs units appeared to track well with the diurnal ozone
variations recorded by the FEM instrument (Figures 1.23-1.24).
18
-------
45
CSAM vs FEM (Ozone)
—FEM —Unit 401 —Unit 402 -—Unit403 —Unit404
0
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.23. Time-series plot of ozone measurements from units #401 through #404 and the FEM
instrument (24-hour average)
CSAM vs FEM (Ozone)
—FEM —Unit 406 —Unit 407 —Unit 408 —Unit 409 —Unit 410
70
60
-O
a 50
c
.o
0
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.24. Time-series plot of ozone measurements from units #406 through #410 and the FEM
instrument (24-hour average)
19
-------
Table 1.11. Correlation coefficient (R2) matrix for the 24-hour average ozone concentrations
measured by the FEM and CSAM units
2
R
FEM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
FEM
1
Unit 401
0.9733
1
Unit 402
0.8260
0.8501
1
Unit 403
0.9554
0.9783
0.8859
1
Unit 404
0.8544
0.8862
0.9388
0.8745
1
Unit 406
0.8312
0.8585
0.8196
0.9024
0.7404
1
Unit 407
0.9420
0.9620
0.8584
0.9826
0.8385
0.9500
1
Unit 408
0.9665
0.9848
0.8338
0.9803
0.8382
0.8940
0.9812
1
Unit 409
0.8042
0.8257
0.8948
0.8877
0.7868
0.9540
0.9206
0.8457
1
Unit 410
0.9603
0.9847
0.8916
0.9862
0.9170
0.8996
0.9772
0.9747
0.8903
1
As noted above in the discussion section for ambient PM2.5 (section 1.3.1), CSAM unit #405 was not included
in this comparison due to erroneous time stamp readings.
Overall, the ozone measurements from the nine CSAMs were very reliable, with data recovery close to 100%
for eight of the nine units tested, which were characterized by modest intra-model variability. Only CSAM
unit #402 had a relatively low data recovery (77.5%). Ozone sensor data for the nine CSAM units (#401,
#402, #403, #404, #406, #407, #408, #409, and #410) showed an excellent correlation (R2: 0.80 - 0.97)
with the corresponding reference measurements collected using a substantially more expensive FEM
instruments (i.e., Thermo 49i) at 5-minute, 1-hour, 8-hour, and 24-hour time resolutions collectively. The
nine CSAM units underestimated the 5-minute average ozone concentrations by 10 to 50% as measured
by the FEM instrument.
1.2.3. Ambient Temperature
Data validation and recovery
Standard QA/QC procedures were used to validate the collected data from the temperature sensor (model
AM 2315, Adafruit) in the CSAM units. Obvious outliers, negative values, and invalid data-points were
eliminated from the data-set. The data recovery for temperature from the nine CSAM temperature sensors
was greater than 99.2% with the exception of unit #402 (77.5%). Descriptive statistics for the nine units and
the RIVR weather station sensor are presented in table 1.12 below.
Table 1.12. Descriptive statistics for temperature data in the nine CSAMs and RIVR station
T (Celsius)
FRM
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
mean
14.4
16.5
15.5
16.1
16.2
16.1
15.9
16.1
16.3
16.3
median
13.5
14.8
14.5
14.7
14.9
14.8
14.6
14.3
14.4
14.4
STDEV
5.5
7.1
5.9
6.9
6.7
6.6
6.6
7.6
7.8
7.6
Count
16368
16300
12733
16298
16303
16305
16293
16301
16304
16300
Recovery {%)
99.7
99.3
77.5
99.2
99.3
99.3
99.2
99.3
99.3
99.3
20
-------
Temperature; infra-model variability
Very low measurement variations were observed among the nine CSAM temperature sensors for ambient
temperature (Celsius) measurements. Figure 1.25 shows the intra-model variability (mean and median
values) for the nine CSAM temperature sensors.
Temperature
£ ¦ mean ± STDEV ¦ median
3
1 30
a
Figure 1.25. Intra-model variability for nine CSAM temperature sensors
CSAM vs RIVR (Temperature; 5-minute mean)
All 5-minute average CSAM temperature measurements from the nine sensors correlated very well with
the corresponding RIVR weather station sensor (model HygroClip2-S3, Rotronic AG) data (R2 > 0.80)
(Table 1.13). The nine CSAM units appeared to track very well with the diurnal temperature variations
recorded by the RIVR weather station sensor (Figures 1.26-1.27).
CSAM vs RIVR (Temperature)
— RIVR —Unit 401 —Unit 402 Unit 403 —Unit 404
50
45
~40 I I
=
« 35 I
o 1
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.26. Time-series plot of temperature measurements from units #401 through #404 and the
weather station sensor (5-minute average)
21
-------
CSAM vs RIVR (Temperature)
—RIVR —Unit 406 —Unit 407 Unit 408 —Unit 409 —Unit 410
50
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.27. Time-series plot of temperature measurements from units #406 through #410 and the
weather station sensor (5-minute average)
Table 1.13. Correlation coefficient (R2) matrix for the 5-minute average temperature values measured
by the weather station sensor and CSAM units
RIVR
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
RIVR
1
Unit 401
0.8993
1
Unit 402
0.8827
0.9964
1
Unit 403
0.9095
0.9959
0.9959
1
Unit 404
0.9152
0.9953
0.9970
0.9989
1
Unit 406
0.9199
0.9922
0.9918
0.9977
0.9975
1
Unit 407
0.9145
0.9903
0.9907
0.9955
0.9960
0.9982
1
Unit 408
0.8312
0.9599
0.9560
0.9581
0.9572
0.9596
0.9612
1
Unit 409
0.8034
0.9506
0.9389
0.9487
0.9451
0.9497
0.9572
0.9771
1
Unit 410
0.8192
0.9635
0.9524
0.9589
0.9570
0.9589
0.9652
0.9662
0.9842
1
CSAM vs RIVR (Temperature; 1-hour mean)
All 1-hour average temperature measurements from the nine CSAMs correlated very well with the
corresponding RIVR weather station data (R2 > 0,81) (Table 1.14). The nine sensor units appeared to track
very well with the diurnal temperature variations recorded by the RIVR weather station (Figures 1.28-1.29),
22
-------
CSAM vs RIVR (Temperature)
— RIVR —Unit 401 —Unit 402 Unit 403 —Unit 404
50
45
0
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.28. Time-series plot of temperature measurements from units #401 through #406 and the
weather station sensor (1-hour average)
CSAM vs RIVR (Temperature)
RIVR —Unit 406 —Unit 407 Unit 408 —Unit 409 —Unit 410
50
45
0
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.29. Time-series plot of temperature measurements from units #406 through #410 and the
weather station sensor (1-hour average)
23
-------
Table 1.14. Correlation coefficient (R2) matrix for the 1-hour average temperature values measured
by the RIVR weather station sensor and CSAM units
2
R
RIVR
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
RIVR
1
Unit 401
0.9047
1
Unit 402
0.8902
0.9982
1
Unit 403
0.9148
0.9971
0.9969
1
Unit 404
0.9200
0.9967
0.9976
0.9993
1
Unit 406
0.9246
0.9935
0.9929
0.9981
0.9979
1
Unit 407
0.9192
0.9921
0.9917
0.9963
0.9965
0.9986
1
Unit 408
0.8428
0.9692
0.9657
0.9673
0.9654
0.9680
0.9690
1
Unit 409
0.8129
0.9566
0.9449
0.9544
0.9508
0.9551
0.9617
0.9820
1
Unit 410
0.8297
0.9702
0.9599
0.9654
0.9631
0.9655
0.9717
0.9753
0.9892
1
CSAM vs RIVR (Temperature; 24-hour mean)
All 24-hour average CSAM temperature measurements from the nine sensors correlated very well with
the corresponding RIVR weather station data (R2 > 0.88) (Table 1.15). The nine sensor units appeared to
track very well with the diurnal temperature variations recorded by the RIVR weather station sensor and also
to be quite accurate (Figures 1.30-1.31).
CSAM vs RIVR (Temperature)
— RIVR —Unit 401 —Unit 402 —Unit 403 —Unit 404
50
45
40
1A
0
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.30. Time-series plot of temperature measurements from units #401 through #404 and the
RIVR weather station sensor (24-hour average)
24
-------
CSAM vs RIVR (Temperature)
—RIVR —Unit 406 —Unit 407 —Unit 408 —Unit 409 —Unit 410
50
45
_40
t/i
3
u 35
u
0
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.31. Time-series plot of temperature measurements from units #406 through #410 and the
RIVR weather station sensor (24-hour average)
Table 1.15. Correlation coefficient (R2) matrix for the 24-hour average temperature measurements
taken by the RIVR weather station sensor and CSAM units
R2
RIVR
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
RIVR
1
Unit 401
0.9790
1
Unit 402
0.8842
0.9044
1
Unit 403
0.9818
0.9982
0.9110
1
Unit 404
0.9828
0.9979
0.9110
0.9998
1
Unit 406
0.9842
0.9970
0.9109
0.9996
0.9997
1
Unit 407
0.9837
0.9966
0.9104
0.9994
0.9996
0.9998
1
Unit 408
0.9764
0.9950
0.8969
0.9928
0.9930
0.9913
0.9916
1
Unit 409
0.9713
0.9919
0.8910
0.9894
0.9897
0.9878
0.9887
0.9991
1
Unit 410
0.9781
0.9954
0.8985
0.9937
0.9940
0.9927
0.9931
0.9993
0.9989
1
1.2.4. Ambient Relative Humidity
Data validation and recovery
Standard QA/QC procedures were used to validate the collected data from the relative humidity sensor
(model AM 2315, Adafruit) in the CSAM units. Obvious outliers, negative values, and invalid data-points were
eliminated from the data-set. The data recovery for relative humidity from the nine of the CSAM relative
humidity sensors was greater than 99.2%, whereas it was 77.5% for unit #402. Descriptive statistics for the
nine units and the RIVR weather station sensor are presented in Table 1.16 below.
25
-------
Table 1.16. Descriptive statistics for the Relative Humidity sensors for the CSAM units and the RIVR
station
RH {%)
RIVR
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
mean
54.3
44.5
44.5
40.8
46.8
45.8
42.3
45.4
38.8
49.3
median
54.9
42.8
46.4
38.0
44.1
44.6
39.0
43.0
36.4
43.8
STDEV
28.7
25.8
22.5
23.8
27.8
25.5
24.5
27.2
26.6
30.2
count (#)
16369
16300
12733
16296
16303
16305
16296
16300
16304
16298
recovery
99.7
99.3
77.5
99.2
99.3
99.3
99.2
99.3
99.3
99.2
Relative Humidity; intra-model variability
Very low measurement variations were observed among the nine CSAM sensor units for relative humidity
(%) measurements (Figure 1.32).
Relative Humidity
mean ± STDEV ¦ median
ll IIII ll II
. tJ> -
vS- \S .N
o* o* ^
Figure 1.32. Intra-model variability for the nine CSAM RH sensors evaluated in this study
CSAM vs RIVR (RH; 5-minute mean)
All 5-minute average relative humidity measurements from the nine CSAMs correlated very well with the
corresponding RIVR weather station data (R2 > 0.91) (Table 1.17). The nine sensor units appeared to track
very well with the diurnal relative humidity variations recorded by the RIVR weather station sensor (Figures
1.33-1.34).
26
-------
CSAM vs RIVR (Relative Humidity)
-RIVR —Unit 401 —Unit 402 Unit 403 —Unit 404
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.33. Time-series plot of RH measurements from units #401 through #404 and the RIVR
weather station sensor (5-minute average)
100
CSAM vs RIVR (Relative Humidity)
—RIVR —Unit 406 —Unit 407 Unit 408 —Unit 409 —Unit 410
11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.34. Time-series plot of RH measurements from units #406 through #410 and the RIVR weather
station sensor (5-minute average)
27
-------
Table 1.17. Correlation Coefficient (R2) matrix for the 5-minute average relative humidity values
measured by the weather station sensor and CSAM units
2
R
RIVR
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
RIVR
1
Unit 401
0.9623
1
Unit 402
0.9559
0.9973
1
Unit 403
0.9598
0.9977
0.9974
1
Unit 404
0.9619
0.9969
0.9957
0.9975
1
Unit 406
0.9628
0.9970
0.9959
0.9959
0.9947
1
Unit 407
0.9610
0.9963
0.9968
0.9979
0.9970
0.9965
1
Unit 408
0.9398
0.9868
0.9852
0.9877
0.9857
0.9866
0.9856
1
Unit 409
0.9192
0.9798
0.9796
0.9821
0.9789
0.9794
0.9801
0.9908
1
Unit 410
0.9270
0.9839
0.9828
0.9868
0.9855
0.9828
0.9870
0.9856
0.9902
1
CSAM vs RIVR (RH; 1-hour mean)
All 1-hour average relative humidity measurements from the nine CSAMs correlated very well with the
corresponding RIVR weather station data (R2 > 0.92) (Table 1.18). The nine CSAM units appeared to track
very well with the diurnal relative humidity variations recorded by the RIVR weather station sensor
(Figures 1.35-1.36).
CSAM vs RIVR (Relative Humidity)
— RIVR —Unit 401 —Unit 402 —Unit 403 —Unit 404
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.35. Time-series plot of RH measurements from units #401 through #404 and the weather station
sensor (1 -hour average)
28
-------
CSAM vs RIVR (Relative Humidity)
—RIVR —Unit 406 —Unit 407 Unit 408 —Unit 409 —Unit 410
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
Figure 1.36. Time-series plot of RH measurements from units #406 through #410 and the weather station
sensor (1-hour average)
Table 1.18. Correlation Coefficient (R2) matrix for the 1 -hour average RH values measured by the
RIVR weather station sensor and CSAM units
2
R
RIVR
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
RIVR
1
Unit 401
0.9671
1
Unit 402
0.9621
0.9981
1
Unit 403
0.9653
0.9981
0.9977
1
Unit 404
0.9671
0.9972
0.9961
0.9977
1
Unit 406
0.9679
0.9973
0.9966
0.9961
0.9948
1
Unit 407
0.9663
0.9967
0.9972
0.9980
0.9972
0.9967
1
Unit 408
0.9462
0.9884
0.9875
0.9894
0.9873
0.9880
0.9871
1
Unit 409
0.9258
0.9813
0.9813
0.9835
0.9803
0.9807
0.9813
0.9918
1
Unit 410
0.9336
0.9853
0.9847
0.9881
0.9867
0.9841
0.9884
0.9873
0.9918
1
CSAM vs RIVR (RH; 24-hour mean)
A!! 24-hour average relative humidity measurements from the nine CSAMs correlated very well with the
corresponding RIVR weather station data (R2 > 0.98) (Table 1.19). The nine CSAMs units appeared to track
very well with the diurnal relative humidity variations recorded by the RIVR weather station sensor
(Figures 1.37-1.38).
29
-------
10
0
11/9/16 11/17/16 11/25/16 12/3/16
12/11/16 12/19/16 12/27/16 1/4/17
80
CSAM vs RIVR (Relative Humidity)
— RIVR —Unit 401 —Unit 402 —Unit 403 —Unit 404
Figure 1.37. Time-series plot of RH measurements from units #401 through #404 and the RIVR weather
station sensor (24-hour average)
o
11/9/16 11/17/16 11/25/16 12/3/16 12/11/16 12/19/16 12/27/16 1/4/17
CSAM vs RIVR (Relative Humidity)
—RIVR —Unit 406 —Unit 407 —Unit 408 —Unit 409 —Unit 410
Figure 1.38. Time-series plot of RH measurements from units #406 through #410 and the RIVR weather
station sensor (24-hour average)
30
-------
Table 1.19. Correlation coefficient (R2) matrix for the 24-hour average RH values measured by the
RIVR weather station sensor and CSAM units
R2
RIVR
Unit
401
Unit
402
Unit
403
Unit
404
Unit
406
Unit
407
Unit
408
Unit
409
Unit
410
RIVR
1
Unit 401
0.9899
1
Unit 402
0.9849
0.9926
1
Unit 403
0.9916
0.9994
0.9928
1
Unit 404
0.9881
0.9983
0.9900
0.9979
1
Unit 406
0.9930
0.9978
0.9925
0.9977
0.9962
1
Unit 407
0.9906
0.9980
0.9923
0.9982
0.9976
0.9985
1
Unit 408
0.9878
0.9965
0.9893
0.9968
0.9954
0.9946
0.9940
1
Unit 409
0.9837
0.9952
0.9887
0.9953
0.9944
0.9926
0.9934
0.9979
1
Unit 410
0.9854
0.9963
0.9897
0.9961
0.9973
0.9948
0.9974
0.9942
0.9963
1
1.3. QA Summary for Co-location Testing
Throughout the CSAM co-location evaluation period in the field, the ozone and particle instrumentation at
the monitoring station was maintained and operated according to standard operating procedures (SOPs) by
the Atmospheric Measurements Branch (AM) at SCAQMD. The Thermo 49i FEM ozone instrument
underwent routine weekly maintenance checks by the station operator to ensure that the instrument was
operating according to specifications. The ozone instrument was challenged with daily precision and weekly
span checks to verify that the instrument was operating within 7% of the expected value during the
challenge. During the co-location testing, the ozone instrument's precision and span checks were all within
7% of their expected values. The ozone instrument at Rubidoux AMS was calibrated on April 21, 2016 and
then again on November 4, 2016. The calibration on November 4, 2016 was an "as-is" equals "final"
calibration, which indicates that the slope and intercept remained the same on the instrument. The level of
routine maintenance, regular calibrations, and daily precision checks assured that the instrument was
collecting quality data that could be submitted to the EPA for regulatory NAAQS attainment determinations.
The Metone FEM PM2.5 BAM 1020 received monthly maintenance, service, and verification of the flow,
temperature, pressure, and whether there were leaks. Throughout the course of the co-location, the flow,
temperature, pressure, and leak checks were all found to be within acceptable limits. The Metone BAM
automatically performed a span check each hour with a reference membrane to ensure that the instrument
was not drifting over time. The calibration on the BAM was performed on June 24, 2016.
The Grimm 180 EDM underwent monthly verifications during the study that involved verifying the
instrument's flow, temperature, and RH values against the reference grade flow and temperature meter.
The unit was calibrated by the manufacturer on November 29, 2016. The combined maintenance,
calibrations, verifications, and checks on the FEM equipment provided assurance that the instrumentation
was operating in good condition and producing quality data.
The CSAM units were checked weekly to ensure proper data logging and reasonability of values. Having ten
CSAM units in one area provided the opportunity to visibly spot check the units for reasonability with each
other in regard to PM2.5 and ozone concentrations. No flow checks were performed during the study on the
Alphasense OPC sensors. Notice was given to spot check the date and time stamp of the CSAM units weekly
31
-------
to compare them with the station time to ensure that the two data sources could be time aligned properly
in the data analysis phases of the project.
32
-------
2. Laboratory Chamber Evaluation
2.1. Methods
A characterization chamber system, designed by the AQ-SPEC team and developed and integrated in
collaboration with AmbiLabs (Warren, Rl), was used for the laboratory testing of two representative CSAM
sensor pods (Units #406 and #409). The chamber system (Figure 2.1) and the methods developed to
evaluate sensor performance under controlled conditions are described in detail in the paper
"Development of an environmental chamber for evaluating the performance of low-cost air quality
sensors under controlled conditions" (Papapostolou et al., 2017). Specifically, the following chamber
system components were used for testing the CSAM sensor pods under controlled environmental
conditions.
• A T/RH controlled rectangular-shaped stainless-steel enclosure [38" (width) x 38" (height) x 54"
(depth)] (referred to as the "outer chamber"), used to conduct aerosol sensors testing experiments
• A T/RH controlled cylindrical-shaped Teflon-coated stainless-steel enclosure [12" (base radius) x
15" (height)] (referred to as the "inner chamber"), used to conduct gaseous sensors' testing
experiments
• A dry, gas- and particle-free air ("zero-air") generation system
• A dust monitor by GRIMM (model EDM180, Ainring, Germany): The EDM 180 spectrometer
provides high-resolution real-time aerodynamic measurements of PMio, PM2.s, PMi.o, TSP, and
PMcoarse particles. The EDM 180 measures light-scattering and is designated as a class III equivalent
method EQPM-0311-195 for ambient PM2.5 mass concentration by the U.S. EPA.
• An aerodynamic particle sizer (APS) byTSI (Model 3321): The APS 3321 spectrometer provides high-
resolution, real-time aerodynamic measurements of particles from 0.5 to 20 urn in diameter. The
APS measures light-scattering intensity in the equivalent optical size range of 0.37 to 20 urn (BAT
for measuring particle size distribution above 0.5 urn in aerodynamic diameter)
• An aerosol generator made by PALAS (model AGK 2000; Karlsruhe, Germany), used to produce
ultrafine and fine particles from various solutions and suspensions
• An O3 analyzer by American Ecotech (model Serinus 10, Warren, Rl). The Serinus 10 is designated
as an equivalent method EQOA-0809-187 by the U.S. EPA (40 CFR Part 53).
• A dynamic dilution calibrator with an internal ozone generator by Teledyne API (model T700U,
San Diego, CA)
• A certified 48.6 ppm nitric oxide cylinder (Air Liquide, Santa Fe Springs, CA)
• An American Ecotech Serinus 40 oxides of nitrogen analyzer (FRM for ambient NO2concentration)
• An HMM100 humidity module by Vaisala to measure temperature and relative humidity
A custom-developed software was used to automatically and remotely control and operate the chamber
and all reference instruments. This software allows for the design of extensive sensor testing
experiments using programmed sequences.
33
-------
Coarse Particle
Dispenser
House Air
Zero Air
Generator
Mixing Duct
Ultrafine/fine
particle
generator
Outer Chamber
Inner [
Chamber [
Communication Plate
x T, RH, Pressure Sensors
Gas Dilution
Calibrator
Manifold
Certified Cylinders
Particle
Instruments
CO
Figure 2.1. Schematic of AQ-SPEC's environmental chamber
Both the outer and inner chambers are capable of maintaining a wide range of temperature and relative
humidity conditions. The HVAC heating/cooling system is used to control the temperature, and a
humidifier/de-humidifier is used to control varying relative humidity. A set of two fans, installed in the
rear wall of the outer chamber and behind the upper wall perforations, generates a circular airflow in the
outer chamber, with the air flowing in from the bottom first through the cooling and dehumidifying coils
and then passing through the heating elements. This air movement mechanism provides for uniform mixing
inside the outer chamber. To achieve target T/RH conditions in the inner chamber, a pump draws the
conditioned air from the outer chamber at a set flow (approximately 10 LPM). This relatively high flow
is also used to mix the air in the inner chamber, creating a homogenous atmosphere. The pump flow is
controlled by a mass flow controller.
Due to their size and the inner chamber volume of 110 L, only two CSAM sensor pods (as opposed to three
units per the standard AQ-SPEC laboratory testing protocol) could be installed inside the inner chamber
for ozone testing. Thus, it was decided to use only two representative units (#406 and #409) for chamber
evaluation for both the PM and gas sensors testing.
Figure 2.2 shows the CSAM PM2.5 and ozone sampling ports as well as those of the reference instruments.
For PM2.5 testing, the rectangular-shaped stainless-steel "outer" chamber was used. For ozone testing, the
CSAM pods were placed inside a smaller cylindrical-shaped Teflon coated stainless-steel unit installed
within the "outer" chamber.
34
-------
Figure 2.2. Two CSAM units installed on the inner chamber base
The CSAM sensor pod reports data at a 5-minute time resolution. The AQ-SPEC testing procedures were
modified accordingly to ensure a representative number of data points were collected for statistical analysis.
A detailed description of the procedures used for testing the PM2.5 and ozone sensors of the CSAM pods is
discussed below.
2.1.1. PM2.5 testing procedure
As noted above, the chamber equipment used to conduct PM2.5 sensor testing in the CSAM sensor pod
consisted of an "outer" chamber, a PALAS particle generator, a FEM GRIMM EDM 180 dust monitor, and
an TSI APS 3321 for measuring aerosol size distribution.
Before beginning an experiment, the chamber wall surfaces were cleaned with Kimwipes dampened in
deionized water. The two CSAM units that were selected for the chamber evaluation, #406 and #409, had
been used previously in the field co-location studies with the other CSAM units. Units #406 and #409 were
mounted on a customized aluminum shelf (Figure 2.2). The sensors' inlets were located approximately
three inches from the reference instrument sampling inlets to ensure that all FRM/FEM monitors were
sampling from approximately the same location inside the chamber. Subsequently, the chamber door was
closed, and the chamber was flushed with dry, particle- and gas-free air. Particle mass and number
concentrations were monitored using the chamber PM reference instruments (GRIMM and APS) to
confirm that the PM mass and count concentration levels inside the chamber were negligible.
The CSAM units logged data locally on an internal SD memory card, and the data were downloaded to a
computer at the conclusion of an experiment. The procedure for evaluating the CSAM pod's ability to
measure PM2.5 involved several concentration ramping experiments under (1) "average ambient" T/RH and
(2) extreme T/RH combinations.
35
-------
Phase 1: Concentration ramping experiments under "average ambient" T and RH conditions
Once the chamber had reached the desired average ambient conditions of 20 °C and 40% relative humidity,
a concentration ramping experiment was initiated. Per the AQ-SPECSensors Lab Testing Protocol, a total of 5
concentration steps were selected to mimic a diverse pollutant profile from very low (0-10 i-ig/m3) to very
high (300 |-ig/m3) PM mass concentrations of ultrafine/fine particles. Each concentration ramping step
was maintained for 210 minutes to allow for at least a steady-state period of 60 minutes. The test
aerosol was created using a 21.9% potassium chloride (KCI) aqueous solution in the PALAS system.
Experimental parameters, such as frequency and duration of aerosol injection into the chamber, were
pre-determined and programmed in the software sequence scheduler. The chamber fans that were used
to create uniform mixing inside the outer chamber were maintained at a constant speed (e.g., frequency
of 25 Hz).
Information derived from this experiment was used to evaluate the sensors' performance parameters, such
as linear correlation coefficient, accuracy, and intra-model variability.
Phase 2: Concentration ramping experiments under extreme T and RH conditions
Four T-RH combinations were used to test the PM2.5 sensor in the CSAM units. For each condition, a 3-
step concentration ramping experiment was conducted. The concentrations were selected to represent low,
medium, and high ambient PM concentrations.
The information derived from these experiments was used to evaluate parameters such as the precision
and effect of T and RH.
Table 2.1. Four representative T-RH combinations
Condition
Temperature (°C)
Relative Humidity (%)
Environment
1
5
15
cold, dry
2
5
65
cold, humid
3
35
15
hot, dry
4
35
65
hot, humid
2.1.2. Ozone testing procedure
The equipment used to test the ozone sensor in the CSAM units consisted of the cylindrical-shaped Teflon
coated stainless-steel inner chamber, a Teledyne T700U gas dilution calibrator and ozone generator, a
certified 48.6 ppm NO cylinder, a Serinus 10 ozone monitor, and a Serinus 40 oxides of nitrogen analyzer.
The procedure for evaluating the CSAM ozone sensor's performance included the following steps: (1)
ozone concentration ramping experiment under "average ambient" T and RH; (2) concentration ramping
experiment at more extreme T and RH; and, (3) an evaluation of potential interferences due to the
presence of NO2.
Phase I: Concentration ramping experiments under ambient T and RH conditions
Once the inner chamber reached the "average ambient" conditions of 20 °C and 40% RH, an ozone
concentration ramping experiment was initiated. A total of 5 concentration steps were selected to mimic a
diverse pollutant profile from very low (0-30 ppb) to very high (300 ppb) ground-level ambient ozone
36
-------
concentrations. Each concentration ramping step was maintained for 150 minutes to allow for a steady-
state period of at least 60 minutes. Ozone was generated using the T700U's internal ozone generator, which
was then introduced to the inner chamber and mixed with conditioned zero air to achieve the desired target
ozone concentration inside the inner chamber. Dilution factors and the expected ozone loss rate had been
taken into account prior to the experiments and were pre-determined and programmed in a sequence. The
information derived from this experiment was used to evaluate parameters such as linear correlation
coefficient, accuracy, and intra-model variability.
Phase II: Concentration ramping experiments under "extreme" T and RH conditions
The same four T-RH combinations that were used for the PM testing were also used for testing the CSAM
ozone sensor's performance (see Table 2.1 above). For each condition, a 3-step concentration ramping
experiment was conducted. The concentrations were selected to represent low, medium, and high ground-
level ambient ozone concentrations. The information derived from these experiments was used to evaluate
parameters such as precision and the effects of T and RH.
Phase III: Potential interference due to the presence of NO2
The effect of an NO2 interferent was evaluated by exposing the two CSAM units to an increasing
concentration of NO2. NO2 was formed by mixing nitric oxide (NO) from a certified cylinder with the ozone
generated in the T700 internal ozone generator inside the T700 dilution calibrator at a ratio of 1.3:1. The NO2
and O3 concentrations were monitored with the reference instruments and the two CSAM units.
2.1.3. CSAM Evaluation Parameters
The following parameters were established for evaluating the CSAM's ability to measure PM2.5 and ozone:
(1) linear correlation coefficient, (2) accuracy, (3) precision, (4) lower detection limit, (5) effect of T/RH,
(6) intra-model variability, (7) data recovery, and (8) effect of interferents (where applicable).
Linear correlation coefficient
This parameter expresses the strength of the linear relationship between the average measurements from
the units tested and the corresponding reference instrument values. The linear correlation coefficient (R2)
was determined in a 5-step pollutant (PM2.5 or ozone) concentration ramping experiment from a very low
to a very high concentration at "average ambient" conditions of 20 °C and 40% relative humidity. Paired
reference instrument and averaged sensor data were entered in an excel spreadsheet, and a linear
correlation coefficient was calculated and reported along with slope and intercept values. An R2
approaching the value of 1 reflects a nearly perfect agreement between the sensors and the reference
instrument readings, whereas a value of 0 indicates a complete lack of correlation.
Accuracy
Accuracy (A) is the degree of closeness between the sensors' measured values and the reference
instrument's value. For the purpose of these chamber tests, accuracy was estimated from the five steady
state periods of the concentration ramping experiment at 20 °C and 40% relative humidity. At each steady
state, accuracy was calculated by:
\X-R\
A(%) = 100 — —n—*100 (1)
37
-------
Where,
X is the average concentration measured by the sensors throughout the steady state period and,
R is the reference instrument average concentration during the same steady state period.
An accuracy value close to 100% implied that the sensors measured exactly what the reference
instrument measured at a specific pollutant concentration. In cases in which sensors overestimated
the reference instrument by more than 100%, the sensor's accuracy was reported as a negative value
as defined in equation (1).
Precision
Precision (P) represents the variation around the mean of repeated measurements of the same pollutant
concentration. This parameter was evaluated at the steady state period of each tested condition. Four T-
RH combinations were used for testing (listed in Table 2.1). In each T/RH combination, a concentration
ramping experiment with low, medium, and high pollutant concentration was performed.
Precision was calculated by data acquired from each steady state concentration ramping step:
P(%) = 100 - SEse™°r * 100 (2)
Where,
sensor is the standard error of the averaged concentrations of the sensors during the steady state period
considered.
X the average concentration measured by the sensors throughout the same steady state period.
Effect of T and RH
The effect of T and RH on sensor's performance are for four the T/RH conditions of "cold, dry", "cold,
humid", "hot, dry", and "hot, humid," as listed in Table 2-1.
Intra-model Variability
Intra-model variability is related to how close the measurements agree among the CSAM units. The intra-
model variability was evaluated through a set of descriptive statistical parameters, such as mean, median,
and standard deviation (each calculated at low, medium, and high pollutant concentrations). These values
were derived from the concentration ramping experiment at 20 °C and 40% RH. For both the aerosol and
ozone experiments, the 5-minute average data from each steady state period were used. For the two CSAM
pod units tested in the environmental chamber, their difference is presented as a percentage and calculated
as follows:
Intra-model variability (%) = Meanhlgh Meaniow ^ ^
Meanaverage
Where,
Meanhigh is the higher value of the two CSAM Pods average concentration at the steady state, Meaniowis the
lower value of the two CSAM Pods average concentration at the steady state, Meanaverage is the average of
the two CSAM Pods average concentration at the steady state.
38
-------
Data recovery
Data recovery is calculated using a percentage ratio of the number of valid sensor data points over the total
number of data points collected during the testing period (e.g., 10 hours of testing at a 5-minute time
resolution results in a total of 120 data points). Data recovery is reported as a percentage and calculated as
follows:
Data recovery(%) =
Nvalid data
100(4)
Ntest period
Where,
Nvalid data is the number of valid sensor data points during the testing period and Ntest period is the total
number of data points for the testing period
Effect of interferents
Metal oxide ozone sensors may be susceptible to nitrogen dioxide cross-sensitivity. In the laboratory, the
effect of NO2 interferent was evaluated by exposing the CSAM units to a ramping experiment of known
NO2 concentration. If CSAM responded to the interferent, a quantitative relationship of units' response
and NO2 concentration was determined.
2.2. Results
2.2.1. Laboratory PM2.5
Linear correlation coefficient: The linear correlation coefficient between the two CSAMs and FEM GRIMM
was determined from the following concentration ramping experiment (Figure 2.3) at 20 °C and 40% RH. As
shown in the figures below, both units #406 and #409 correlated well with the FEM instrument (R2 > 0.99).
Unit 406 recorded higher PM2.5 mass concentrations than unit 409. As a secondary reference, APS 3321
measured similar PM2.5 mass concentrations to those reported by the FEM GRIMM (R2 > 0.99, slope = 1.01,
and intercept = 3.81).
CSAM PM25 vs GRIMM vs APS (20 °C, 40% RH)
—GRIMM —APS Unit 406 Unit 409
PM2 5 mass conc. (ng/m3)
PM2 5 mass conc. (ng/m3)
500
400
300
ra 200
y = 1.40x + 13.03
R2 = 0.9931
100 200 300 400 500
GRIMM
y = 1.30x + 10.08
R2 = 0.9942
100 200 300 400 500
GRIMM
100
Q.
R2
Slope
Intercept
Unit 406 vs GRIMM
0 9931
1.40
13.03
0 1——"
Unit 409 vs GRIMM
09942
1.30
10.08
0
200
400
600
800 APS vs GRIMM
09978
1.01
3.81
Time (minute)
Figure 2.3. PM2.5 mass concentrations as measured by the CSAM and the reference instruments used
for these tests (FEM GRIMM and APS)
Accuracy: The CSAM PM2.sdata showed low to moderate accuracy compared to the FEM GRIMM PM2.5
during the concentration ramping experiment between 0-300 i-ig/m3. With accuracy ranging between 17.1%
and 62.6%. Overall, the two CSAMs overestimated the PM2.5 mass concentrations as measured by the FEM
GRIMM (Table 2.2).
39
-------
Table 2.2. CSAM PM2.5 Accuracy
Steady State
(#)
CSAM mean
(Hg/m3)
FEM GRIMM
(W5/m3)
Accuracy
(%)
1
19.8
13.0
47.6
2
63.1
34.5
17.1
3
138.9
81.8
30.2
4
267.6
178.5
50.1
5
407.2
296.4
62.6
Precision: The CSAMs demonstrated excellent precision (97-99%) in measuring PM2.5 under most of the
conditions tested (Table 2.3), except at 5 °C and 65% relative humidity. The precision of both units was
affected by the low temperature and high humidity (and high humidity variation). The precision was
between 74 and 91% for low to high PM2 5 mass concentrations at 5 °C and 65% RH. At 35 °C and 15% RH,
and also at 35 °C and 65% RH, precision was very high when ranging between low and high PM2.5 mass
concentrations, indicating the two CSAMs demonstrated similar performance under most
environmental conditions.
Table 2.3. CSAM PM2.5 precision under extreme T and RH conditions
5 °C 35 °C
Low CSAM GRIMM CSAM GRIMM
pm2.5
15%
98.9
99.2
99.3
99.7
(Hg/m3)
65%
74.0
93.0
98.8
97.1
5°C
35 °C
Medium
CSAM
GRIMM
CSAM
GRIMM
PM2.5
15%
99.6
99.7
99.7
99.8
(Hg/m3)
65%
86.2
96.7
99.4
99.5
5°C
35 °C
High
CSAM
GRIMM
CSAM
GRIMM
PM2.5
15%
99.6
99.7
99.7
99.5
(Hg/m3)
65%
90.5
97.9
99.7
99.7
Effect of T and RH: The CSAM's ability to measure low to high PM2.5 concentrations was evaluated under
four T-RH conditions (Figure 2.4). At 5 °C and 65% RH, both units #406 and #409 reported spiked PM2.5 mass
concentrations, whereas the FEM GRIMM and APS reported stable concentrations with very low
variability. It is possible that this CSAM response might be associated with the sensor's limitation in
measuring PM2.5 under low temperature and high humidity (also humidity variation).
40
-------
CSAM PM25 vs GRIM M vs APS (5 °C, 15% RH)
—GRIMM —APS —Unit 406 Unit 409
100 200
300 400
Time (minute)
500 600
CSAM PM25 vs GRIMM vs APS (35 °C, 15% RH)
—GRIMM —APS —Unit 406 Unit 409
300 400 500
Time (minute)
CSAM PM25vs GRIMM vsAPS(5 °C, 65% RH)
—GRIMM —APS —Unit 406 Unit 409
) 100 200 300 400 500
Time (minute)
CSAM PM2 5 vs GRIMM vs APS (35 °C, 65% RH)
—GRIMM —APS —Unit 406 Unit 409
-1000
100 200 300 400 500 600 700
Time (minute)
Figure 2.4. Effect of T-RH on CSAM PM2.5 Performance
Data Recovery: Both CSAMs had excellent data recovery (nearly 100%) for PM2.5-
Intra-model Variability: At 20 °C and 40% RH, units #406 and #409 correlated well with each
other (R2 > 0.99, slope = 1.08, intercept = 2.06). Low measurement variations were
observed at low, medium, and high PM2.5 mass concentrations (Figure 2.5).
(a) PM2.5 mass conc. (ng/m3)
500
y = 1.08x + 2.06
R2 = 0.9998
100 200 300 400 500
Unit 409
Low PM2 5 Conc. (20 °C, 40% RH)
¦ mean ¦ median
Medium PM2 5 Conc. (20 °C, 40%RH)
¦ mean ¦ median
(d) High PM2 5 Conc. (20 °C, 40%RH)
— ¦ mean ¦ median
Figure 2.5. (a) Unit #406 vs unit #409; (b-d) intra-model variability at low, medium, and high PM2.5 conc.
41
-------
2.2.2. Laboratory Ozone
Linear correlation coefficient: The linear correlation coefficient between the two CSAM sensor pods and
the FEM ozone was determined from the following concentration ramping experiment (Figure 2.6) at 20
°C and 40% RH. As shown in Figure 2.6 below, the CSAM units followed the ozone concentration
changes, as measured by the FEM ozone monitor. Both units #406 and #409 correlated well with the
FEM instrument (R2 > 0.95). However, the two CSAM units had a high baseline value of 20 ppb (intercept:
~ 20 ppb). At high ozone concentrations, the two CSAM units significantly underestimated the FEM
ozone concentration (slope: 0.12-0.13).
CSAM Ozone vs FEM ozone (20 °C, 40%RH)
—FEM Unit 406 Unit 409
300
O)
C
o
O 200
UD
O
~ 100
Ozone Cone, (ppb)
y = 0.13x + 20.45
R2 = 0.95
300
0)
c
o
O 200
CTi
O
~ 100
Ozone Cone, (ppb)
y = 0.12x+ 19.9
R2 = 0.95
200 400
Time (minute)
800
0 100 200
300
0
100 200 300
FEM ozone
FEM ozone
R2
Slope
Intercept
Unit 406 vs FEM
0.95
0.13
20.45
Unit 409 vs FEM
0.95
0.12
19.90
Figure 2.6. CSAM vs FEM ozone concentration ramping experiment (20 °C, 40% RH)
Accuracy: The CSAMs' accuracy in measuring ozone decreases as the ozone concentration increases, and
ranges from 86.4% to 17.5% (Table 2.4). The CSAM units had baseline reading of approximately 20 ppb and
the highest concentration reported was around 55 ppb, whereas the FEM's measured ozone concentration
was between 0 to 310 ppb. The CSAMs in most cases underestimated the ozone concentrations as measured
by the FEM instrument.
Table 2.4. CSAM Ozone Accuracy
Steady State
CSAM mean
FEM
Accuracy
(#)
(PPb)
(PPb)
(%)
1
20.0
17.6
86.4
2
23.3
32.9
71.0
3
29.9
62.7
47.6
4
45.2
168.1
26.9
5
54.0
308.8
17.5
Precision: The CSAMs demonstrated good precision under most of the environmental conditions
tested for this evaluation (Table 2.5). At 5°C and 15% RH, CSAM unit #406 reached its maximum reading
of 500 ppb when the FEM ozone changed from 100 ppb to 300 ppb. Precision under this condition could
not be evaluated because unit #406 appeared to be showing abnormal measurements. At 5 °C and 65% RH,
unit#406 reported unusually high ozone concentrations when the FEM was recording a stable ozone
concentration around 350 ppb. At 35 °C and 15% and 65% RH, the two CSAMs showed a similar high
precision compared to the ambient conditions.
42
-------
Table 2.5. CSAM Ozone Precision under extreme T and RH conditions
5 °C 35 °C
Low CSAM FEM CSAM FEM
Ozone
15%
99.0
97.6
99.6
99.7
(PPb)
65%
99.7
100.0
99.3
99.5
5 °C 35 °C
Medium CSAM FEM CSAM FEM
Ozone
15%
99.4
99.1
99.6
99.6
(PPb)
65%
99.8
100.0
99.7
99.7
5 °C
35 °C
High
CSAM
FEM
CSAM
FEM
Ozone
15%
N/A
99.7
99.7
99.9
(PPb)
65%
92.9
100.0
99.6
99.9
Effect of T and RH: The CSAMs' ability to measure low to high ozone concentrations was evaluated at the
fourT-RH conditions (see table 2.1 above). Unit #409 was consistent with its performance as it was exposed
to ambient T-RH. However, at high ozone concentrations, unit #406 reported unusually high values and a
maximum reading of 500 ppb at 5 °C and 15% and 65% (Figure 2.7).
CSAM Ozone vs FEM (5 °C, 15% RH)
—FEM Unit 406 Unit 409
CSAM Ozone vs FEM (5 °C, 65% RH)
—FEM Unit 406 Unit 409
600
500
-£3
EL
EL
400
U
C
O
300
u
GJ
C
o
200
O
100
> 200
100
100
200 300
Time (minute)
400
CSAM Ozone vs FEM (35 °C, 15% RH)
—FEM —Unit 406 ~Unit409
100 200 300
Time (minute)
CSAM ozone vs FEM (35 °C, 65% RH)
—FEM —Unit 406 =Unit 409
300
250
3"
a. 200
o 150
u
O)
O 100
o
50
0
200 300
Time (minute)
300
250
-El
EL
EL
200
U
C
o
150
u
0)
c
o
100
O
50
0
-50
200 300
Time (minute)
Figure 2.7. Effect of T-RH on CSAM ozone performance
Data Recovery: Both CSAMs had excellent data recovery (close to 100%) for ozone.
Intra-model Variability: At 20 °C and 40% RH, units #406 and #409 had very high linear correlation between
43
-------
each other (R2 > 0.99, slope = 1.41, and intercept = - 1.41). Low ozone measurement variations were
observed at low, medium, and high ozone concentrations. However, it is notable that at 5 °C and 15% and
65% RH and at high ozone concentrations, unit # 406's performance was significantly different than that of
unit #409, reporting spiked values and a very high concentration of nearly 500 ppb (Figure 2.8).
a.
(a)
60
us 40
o
c
D
20
Ozone Cone, (ppb)
y = l.lOx -1.41
R2 = 0.99
20 40
Unit 409
60
b.
c.
d.
(b)
Ozone Cone. (20 °C, 40% RH)
¦ mean e median
-------
2.3. QA summary for laboratory testing
The GRIMM monitor is serviced and calibrated by the manufacturer annually. The most recent service
and calibration procedures are valid until January 2018. The CSAM evaluation experiments using the
GRIMM were conducted between January 12 and January 27, 2017. The APS is serviced and calibrated
every three years. As the instrument was first used on August 25, 2016, its calibration is valid until August
2019. The Serinus 10 ozone monitor is calibrated every six months (or more often). The most recent
calibration was performed on February 1, 2017. The ozone evaluation experiments performed on the
CSAM units were conducted between February 8 and February 16, 2017. The Serinus 40 oxides of
nitrogen monitor is calibrated every six months (or more often). Its most recent calibration was
performed on October 27, 2016. The NOx interferent experiment was conducted on April 27, 2017.
Prior to the CSAM's laboratory testing, routine cleaning procedures were followed on the chamber's
mixing duct. The mixing duct, shown in Figure 2.1, was fully disassembled and the various parts were
flushed with clean air to remove deposited particles from the wall surfaces.
One deviation from the AQ-SPEC laboratory testing protocol methods was the period of time maintained
each steady state. The AQ-SPEC testing procedures were developed for testing sensors with data averaged
at 1 minute intervals. However, the CSAM unit was designed to report data at a 5-miuten time resolution.
Therefore, the experiments were modified accordingly to ensure that a representative number of data points
were collected for the CSAM PM and ozone sensors' evaluations. The concentration steady-state period in
these experiments was maintained for a longer time period than was typically maintained with other sensor
applications. For example, in the CSAM PM test, the steady-state period was increased from 150 min to 210
min, and in the CSAM ozone test, the steady-state period was increased from 40 to 150 minutes. A detailed
description of the testing procedures used for testing the PM2.5 and ozone sensors in the CSAM units is
discussed in section 2.1.1. and section 2.1.2.
Due to the increased length of each experiment and demanding testing schedule, the sensor's climate
susceptibility was evaluated using four representative weather (T and RH) combinations based on our
objective judgement (Table 2.1.). Although the chamber has the potential to generate much lower or
higher T and RH, we could not afford to conduct exhaustive testing of additional weather combinations.
The four T/RH combinations were evaluated at low, medium, and high pollutant concentrations for a
total of 12 experiments. In routine AQ-SPEC sensor testing, 27 experiments are conducted for variations
in T/RH/pollutant concentrations.
Additionally, as discussed in the previous section, due to the chamber's size limitation, only two CSAM
pods were evaluated rather than the three described in the original AQ-SPEC laboratory testing protocol.
In this study, the CSAM's response time was not tested because of the chamber's system design. After
units #406 and #409 were installed inside the chamber, the chamber door was closed. The pollutant
concentration was then gradually increased over 30 minutes (gas) or 90 minutes (aerosol) to the steady-
state set-points. Therefore, the sensor's response time was not evaluated.
Finally, with respect to the PALAS artificial aerosol generation system, it is possible to approximate a
broad range of PM mass concentrations. However, the chemical nature, the pre-defined size
distribution, and the physical properties of the generated artificial aerosol cannot replicate the diverse
profile of an urban ambient aerosol chemical composition.
45
-------
3. Field Deployment
After the field and laboratory (chamber) testing activities were completed, ten CSAM pods were deployed
throughout the South Coast Air Basin (SCAB) and the Salton Sea Air Basin (SSAB) in Southern California.
The rolling deployment extended for three months, from January 11, 2017 to April 21, 2017. The pods
were clustered in the three distinct geographical areas of Long Beach, Jurupa Valley, and Coachella
Valley. In each of these areas, three locations were chosen as deployment sites. Each of these three distinct
geographical areas exhibits its own unique air quality challenges. Specifically, the Long Beach area included
a near-roadway location and two environmental justice locations that are affected by their proximity to
localized sources that include refineries, railyards, roadways, and other industrial facilities. The close
proximity to these sources in the Long Beach area causes a number of environmental justice
problems in the region. Due to the common occurrence of daytime on-shore winds pushing ozone pre-
cursors inland, ozone is not typically a significant concern for communities located near the shore.
The Jurupa Valley and Coachella Valley are both inland locations that are impacted by elevated ozone and
particulate matter concentrations. In regard to the National Ambient Air Quality Standard (NAAQS)
attainment, the SCAB was designated with a nonattainment ("extreme") status for its 2008 8-hour ozone
standard (0.075 ppm) and designated with a nonattainment ("serious") status for the 2006 24-hour PM2.5
standard (35 ng/m3). The only two SCAQMD's stations with designated values over the 24-hour PM2.5 NAAQS
within the SCAB in the most recent three-year period (2013-2015) were the Riverside/Rubidoux and the
Mira Loma Air Monitoring stations. In Riverside County in 2015, 76 days exceeded the 2015 8-jhur ozone
standard (0.070 ppm).
The SSAB was designated with as nonattainment ("Severe-15") status for the 2008 8-hour ozone standard
(0.075 ppm) and was designated as unclassifiable/attainment for the 2006 24-hour PM2.5 standard (35
Hg/m3). The area was previously in non-attainment status for PM10 due to mechanically generated dust
from agricultural activities, but it is expected to meet the current PM10 standard pending documentation
submittal and subsequent U.S EPA approval. The Coachella Valley region had 58 days in 2015 that exceeded
the 2015 8-hour ozone National Ambient Air Quality Standard (NAAQS) at 0.070 ppm (2016 SCAQMD
AQMP). For these reasons, these locations were selected as deployment sites for the CSAM pods.
The field deployment from January 11 to April 21, 2017 experienced varied weather conditions among the
three locations. The deployment period coincided with a record-setting rainy season in southern California.
Long Beach experienced 23 rain days and 8 fog days for the duration of the study. The 23 rain days
accounted for a total of 12.8 inches of rain. The temperature ranged from 3.9 to 30 °C with a mean of 15.6
°C. RH ranged from 6-100% with an average of 64% RH. The Jurupa Valley experienced 22 rain days and 4
fog days. The 22 rain days accounted for a total of 5.2 inches of rain. The temperature ranged from 0 to 33
°C with an average temperature of 15.6 °C. The RH ranged from 11-100% with an average RH of 63%. The
Coachella Valle experienced 12 rain days and 1 fog day. The 12 rain days accounted for a total of 1.25 inches
of rain. The temperature ranged from 1.1 to 37.8 °C with an average temperature of 18.3 °C. The RH ranged
from 3-100% with an average of 46.6%. (Source: Weather Underground
https://www.wunderground.com/history/).
3.1. Methods
The CSAM units were deployed in a rolling fashion at nine different locations. Seven of the pods were
placed at existing SCAQMD air monitoring stations (AMS) including a duplicate unit at the Rubidoux AMS.
46
-------
The three remaining CSAM pods were deployed at Ventura Transfer Company, Valley View Elementary
School, and the Jurupa Area Recreation and Parks District (JARPD). Table 3.1 describes the nine field
deployment locations, geographical coordinates (latitude and longitude), and the rationale for the site
selection. Figure 3.1 shows the location of the sensors (colored dots) within the South Coast Air Basin
(outlined in blue).
Table 3.1. CSAM Deployment Locations
Site Name
Pod#
Latitude/ Longitude
Address
Rationale
Hudson AMS
401
33°48'8.82"N
118°13'11.78"W
2425 Webster St. Long Beach,
CA 90806
Near-road/High
Concentration/EJ
Area
710 NR AMS
402
33°51'34.73"N
118°12'2.64"W
5895 Long Beach Blvd. Long
Beach, CA 90810
Near-road/High
Concentration/EJ
Area
Ventura Transfer
Company
404
33°48'57.62"N
118°13'57.87"W
2418 E 223rd St.
Carson, CA 90810
Near-source/High
Concentration/EJ
Area
Rubidoux AMS
(Primary)
403
33°59'58.57"N
117°24'57.88"W
5888 Mission Blvd.
Riverside, CA 92509
High
Concentration
Rubidoux AMS
(Collocated)
405
33°59'58.57"N
117°24'57.88"W
5888 Mission Blvd.
Riverside, CA 92509
High
Concentration
Jurupa Area
Recreation and Parks
District
407
34° 0'5.89"N
117°28'26.35"W
4810 Pedley Rd. Jurupa
Valley, CA 92509
High
Concentration
Mira Loma AMS
408
33°5946.98"N
117°29'32.67"W
5130 Poinsettia PI.
Riverside, CA 92509
High
Concentration
Saul Martinez AMS
406
33°34'19.27"N
116° 349.86"W
65705 Johnson St.
Mecca, CA 92254
High Concentration/
EJ Area
Valley View
Elementary School
409
33°40'1.08"N
116°10'39.07"W
85-270 Valley Rd.
Coachella, CA 92236
High Concentration/
EJ Area
Indio AMS
410
33°42'30.70"N
116°12'55.57"W
46990 Jackson St.
Indio, CA 92201
High
Concentration
47
-------
CSAM Pod Site Locations
Figure 3.1. Sensor Pod Placement in the South Coast Air Basin
48
-------
The rolling deployment was staggered to accommodate access to both the sites and the sensor pods. Table 3.2
shows the deployment dates and the total number of days and weeks each CSAM pod was deployed.
CSAM PODs #406 and #409 were delayed in their initial deployment because they were undergoing AQ-
SPEC laboratory evaluations. Unit #404 at Ventura Transfer Company experienced date and time stamp
issues that resulted in an invalidation of all data collected after March 17, 2017.
Table 3.2 Deployment Location and Dates
Pod#
Site Name
Start Date
End Date
# of Days
# of Weeks
401
Hudson AMS
1/13/2017
4/19/2017
96
13.7
402
710 NR AMS
1/13/2017
4/15/2017
92
13.1
403
Rubidoux AMS (Primary)
1/11/2017
4/15/2017
94
13.4
404
Ventura Transfer Company*
1/17/2017
3/17/2017
59
8.4
405
Rubidoux AMS (Collocated)
1/11/2017
4/15/2017
94
13.4
406
Saul Martinez AMS**
2/22/2017
4/19/2017
56
8.0
407
Jurupa Area Rec and Parks District
1/20/2017
4/21/2017
91
13.0
408
Mira Loma AMS
1/11/2017
4/15/2017
94
13.4
409
Valley View Elementary**
2/22/2017
4/19/2017
56
8.0
410
Indio AMS
1/17/2017
4/19/2017
92
13.1
* Date/time issues starting March 27, 2017
** Delay for laboratory testing
After initial deployment of the CSAM pods, bi-weekly to monthly site visits were conducted to ensure
proper operation, record the date/time, and copy the data from the SD card. Information on all site visits
along with repairs and pod issues were recorded in the electronic CSAM Field Deployment log book. Figures
3.2- 3.9 below show the locations of CSAM pods #401, #402, #404, #406, #407, #408, #409, and #410.
Figure 3.2. CSAM Pod #401 at Hudson AMS
49
-------
Figures 3.4. CSAM Pod #404 at Ventura Transfer Company
Figures 3.5. CSAM Pod #406 at Saul Martinez Elementary School
Figures 3.6. CSAM Pod #407 at Jurupa Area Recreation and Parks District
50
-------
Figures 3.7. CSAM Pod #408 at Mira Loma AMS
Figures 3.8. CSAM Pod #409 at Valley View Elementary School
Figure 3.9. CSAM Pod #410 at Indio AMS
3.2. Results
Overall, the CSAM units performed well for data recovery with the exception of the particulate matter data
(Table 3.3). Several documented events accounted for reduced data recovery. CSAM #401 at Hudson AMS
experienced a power outage due to a blown ground fault interrupter (GFI). The unit was without power for
approximately three days (January 22, 2017 at 13:35 to January 25, 2017 at 12:40). CSAM #402 at the 710
NR AMS also stopped working for approximately 13 days (March 8, 2017 at 17:20 to March 21, 2017 at 9:20),
during which time the unit displayed VOC rather than ozone concentrations and did not record any data.
The unit resumed collecting valid data after a site visit and a manually powered on/off cycle. CSAM unit #405
did not display T and RH readings consistently during the co-location testing period and was designated to
51
-------
the duplicate CSAM unit at Rubidoux AMS. CSAM unit #408 at Mira Loma AMS experienced a power failure
due to heavy rains penetrating the sealed tote containing the power cord and 12V adaptor. The 12V adaptor
shorted out internally and the unit was without power for approximately eight days (January 24, 2017 at
20:15 to February 1, 2017 at 16:40) while a new power adaptor was sourced and installed.
Table 3.3. Percent Data Recovery
Unit
Location
PM2.5
Ozone
Temp
RH
401
Hudson AMS
92.6%
96.5%
94.0%
94.1%
402
710 NR AMS
85.9%
85.9%
85.9%
85.9%
403
Rubidoux AMS (Primary)
56.2%
99.6%
99.2%
99.2%
404
Ventura Transfer
Company
43.9%
99.6%
99.6%
99.6%
405
Rubidoux AMS
(Collocated)
66.4%
99.6%
43.1%
43.1%
406
Saul Martinez AMS
0.7%
99.6%
99.6%
99.6%
407
JARPD
80.0%
99.6%
97.3%
97.9%
408
Mira Loma AMS
20.9%
91.2%
91.1%
91.0%
409
Valley View Elementary
5.4%
99.6%
99.1%
99.1%
410
Indio AMS
58.9%
99.6%
99.1%
99.2%
The data recovery for T and RH exceeded 97% for all pods except for units #401, #402, #405, and #408. Data
recovery for ozone was greater than 96% for all CSAM except units #402 and #408. The data recovery for PM2.5
varied from 0.7% to 92%. The reduced data recovery for PM2.5 was caused by a known error with the
Alphasense OPC-N2. The OPC-N2 initiates a fault error after experiencing high RH. Once the error is
initiated, the device reads 0.00 i-ig/m3 or near zero until the device is manually powered on/off. The error is
not consistent between devices, or even for the same device, and occurs randomly during high RH conditions.
Unfortunately, this error could not be remedied with the software generated reset on the CSAM unit each
night. The unit did not recover from this error unless it was manually powered off/on during a site visit.
Thus, all values less than 1 i-ig/m3 were considered invalid and removed from the data set. Both units that
were tested in the chamber (#406 and #409) had extremely low data recovery after being deployed in the
field (0.7 and 5.4%, respectively). The next lowest PM2.sdata recovery (20%) was recorded for unit #408 at
the Mira Loma AMS site. This drastic reduction in data recovery for the two units that underwent
laboratory testing was likely to have been caused by the high PM loadings experienced during chamber
testing, which negatively affected the operation of the optic system, thereby resulting in a substantial
number of 5-minute PM2.5 readings less than 1 i-ig/m3.
3.2.1. Ambient PM2.5
The CSAM pods utilize an Alphasense Ltd. OPC-N2 particulate matter optical sensor. Overall, the sensors
correlated well with one another during the co-location period with the exception of unit #401 (Table 3.4). As
a result, the differences noted between among most of the sensors during this field deployment should
be indicative of quantitate differences in PM2.5 levels between different locations. The complexity of
inter-comparing descriptive statistics is increased when there are issues with data recovery and reliable
sensor performance. For example, CSAM pods #406 and #409 (deployed in the Coachella Valley) were
characterized by extremely low data recovery (<10%) for PM2.5 and, therefore, summary statistics from this
pod were not included in our analysis. Because CSAM pod #401 did not correlate well with the reference
instrumentation during the co-location testing, its data are also suspect in regard to providing reliable
52
-------
ambient PM2.5 data during this deployment.
Table 3.4. CSAM PM2.5 Summary Statistics
PM2.5{ng/m3)
Long Beac
1
Jurupa
Valley
Coachella Valley
401
402
404
403
405
407
408
406
409
410
Min
1
2.55
1
1
1
1
1
1.4
1
1
Max
451.1
488.5
279.9
124.9
165.1
456.5
470.6
4.6
7.2
209.0
Median
5.1
15.9
5.0
4.4
7.9
5.2
10.3
1.9
1.7
3.6
Mean
14.6
24.3
7.4
9.2
14.5
11.5
20.2
2.3
2.3
6.5
SE
0.2
0.2
0.2
0.1
0.1
0.2
0.4
0.1
0.0
0.1
STDEV
36.6
29.8
14.8
13.7
19.3
22.7
31.4
0.8
1.3
12.8
Recovery {%)
96.2
85.9
43.9
56.2
66.4
80.0
20.9
0.7
5.4
58.9
Table 3.5. FEM Metone BAM PM2.5 Summary Statistics
FEM PM (ng/m3)
Rubidoux
MLVB
Min
0.0
0.0
Max
47
61
Median
9
11
Mean
10.6
12.6
STDEV
8.0
7.9
SE
0.17
0.17
Recovery (%)
93.2
92.7
Long Beach
The three units deployed in Long Beach included CSAM pods #401, #402, and #404. The time series for
these three units is shown in Figure 3.10. Unit #402 at the 710 NR AMS measured the highest mean
concentration of PM2.5among all nine locations. In regard to the other two Long Beach sites, the mean
concentration at the 710 NR AMS (24.2 i-ig/m3) was more than three times the mean concentration of that
observed at the Ventura Transfer Company (unit #404, 7.4 i-ig/m3) and 1.5 times the mean concentration
at Hudson AMS (unit #401, 14.6 ng/m3). The 710 NR AMS is located in close proximity (~15 miles) to
interstate 710, which is one of the primary goods movement corridor routes going inland from the marine
ports.
53
-------
Long Beach
E
OJD
a.
450
400
350
300
250
200
150
100
50
1/11 1/16 1/21 1/26 1/31 2/5 2/10 2/15 2/20 2/25 3/2 3/7 3/12 3/17 3/22 3/27 4/1 4/6 4/11 4/16 4/21
Figure 3.10 Long Beach PM2.5
The measurements from unit #401 at Hudson AMS provide a good example of how to manage questionable
sensor data. First, unit #401 did not perform well during the co-location period. Moreover, the time series
for the Hudson AMS unit included time periods when sensor response was characterized by
different levels of data quality, as illustrated in Figure 3.11 using a colored scaled from green
(trustworthy-normal response) to orange (suspect-high readings) to red (not-trustworthy-no response/flat
line). From January 13, 2017 to March 2, 2017, the sensor reported PM2.sthat were consistently at
around 3 i-ig/m3, which is interesting for its closeness to the mean of the co-location period (3.2 i-ig/m3;
see co-location testing in Chapter 1 above). Between March 2, 2017 and March 21, 2017, the same sensor
reported high PM concentrations. Unfortunately, due to data recovery issues with the other two sensors in
the Long Beach region duringthis specific period, the results cannot be confirmed to be representative of the
Long Beach area. From March 22,2017 to April 19, 2017, the sensor indicated changes in PM and tracked very
closely to unit #402 at the 710 NR AMS (see the time series in Figure 3.12). This is a good example of how
data from one sensor can be validated by using that from another unit when the two devices are within a
reasonable proximity from each another. Building this type of logic and capability into a real-time cloud-
based application to validate and display sensor data is important in the evolution of low-cost air quality
sensor technology and applications.
500
400
300
200
100
—
Hudson AMS (401)
—
—
—
1/11 1/18 1/25 2/1 2/8 2/15 2/22
3/1 3/8 3/15 3/22 3/29 4/5 4/12 4/19
Figure 3.11. Hudson AMS time series
54
-------
120
Hudson AMS (401) and 710NR (402)
401 402
Figure 3.12. Hudson AMS and 710 NR AMS
Jurupa Valley
Four CSAM units were deployed in the Rubidoux area south of State Route (SR) 60 and west of Interstate 15.
The three locations included the Rubidoux AMS, where unit #403 and #405 (co-located units) were
deployed, Mira Loma AMS, where unit #408 was installed, and Jurupa Area Recreation and Parks District
(JARPD,) where unit #407 was operated. The three locations in the Jurupa Valley are relatively close to one
another, with a maximum distance of 5.4 miles separating the Rubidoux and Mira Loma AMS. The Mira
Loma AMS and JARPD are 1.3 miles apart, with a railway and a four-lane roadway between them. These two-
lane sources could have a direct impact on either location dependent of wind direction. The 24-ouhr mean
concentrations at these four locations are shown in Figure 3.13.
Figure 3.13. Jurupa Valley PM2.5 24-hour time series
Considering the 24-hour averages for the Jurupa Valley, the CSAM units show multiple days with 24-hour
PM2.5 concentrations above the NAAQS of 35 |ig/m3. These high particulate 24-hr average concentrations
that exceed the NAAQS requirements warrant additional scrutiny with higher-cost FEM instrumentation.
Both the Rubidoux and Mira Loma AMS are equipped with FEM BAM PM2.5 instruments. The FEM BAM at
Rubidoux does not indicate a single 24-hour average that exceeds the NAAQS of 35 jig/m3 (Figure 3.14).
55
-------
The overestimation of PM concentrations in a community deployment setting may raise unfounded
concerns over exposure levels in the region.
CUD
3
fM
RIVR FEM BAM and CSAM
- 403 405 RIVR BAM
Figure 3.14. Rubidoux FEM BAM and CSAM PM2.5 time series
The Mira Loma AMS is within 1.3 miles of the JARPD location, where unit #407 indicated several days that
exceeded the NAAQS for PM25, whereas the FEM BAM unit at the Mira Loma AMS did not indicate any
exceedances throughout the study, as shown in Figure 3.15.
E
00
(N
Mira Loma FEM BAM and JARPD CSAM
407 MLVBBAM
1/10 1/17 1/24 1/31 2/7 2/14 2/21 2/28 3/7 3/14 3/21 3/28 4/4 4/11 4/18
Figure 3.15. Mira Loma FEM BAM and JARPD CSAM 24-hour PM2.5 Data
The two units co-located at the Rubidoux AMS tracked well with one another and showed a correlation (R2)
value of 0.62. The mean PM2.5 concentration for unit #403 was 9.2 i-ig/m3, whereas the mean for #405 was
higher, at 14.5 i-ig/m3, The regressions, shown in Figure 3.17, between units #403 and #405 demonstrate
linearity between the two sensors, with a notable time period during which #405 recorded higher
concentrations than #403. Upon further examination of the time series data, it appeared that unit #403 had
begun to underestimate #405 after February 26, 2017. Upon filtering the data to include only values
recorded before February 26, 2017, the R2 between the two units increased from 0.64 to 0.97. It is likely
that the OPC-N2 in the CSAM #403 pod was losing sensitivity to changes in PM concentrations over time.
56
-------
Figure 3.16. Rubidoux Co-located CSAM time series
Rubidoux Co-located Rubidoux Co-located
24-hr Mean y = 1.09l4x+ 5.2674 24-hr Mean Filtered > 2/26/17 v
R2 = 0.6977 70.0 |
70.0
•
*•••
•
tp* ' j
.. * 60.0
... • 60.0 .t E
E ^ 50.0
50.0
^ 40 0
g 30.0
^ 200
Lf»
o
10.0
0.0
0.0 10.0 20.0 30.0 40.0
403 PM2 5(ng/m3)
3
40.0
£ 30.0
« 20 0
o
10.0
0.0
fv4
20.0 30.0 40.0
403 PM2l5(ug/m3)
Figure 3.17. Rubidoux Co-Location Scatterplots Full Time Period (left) and Filtered for January 11, 2017
to February 11, 2017 (right)
CSAM unit #408 at the Mira Loma AMS had the highest mean concentration among all theJurupa Valley
locations. However, the data recovery for this device was low at 20.9%. A meaningful inter-
comparison among unit #408 and the units at other locations in Jurupa Valley is difficult to attain due
to the low data recovery. CSAM #405 (Rubidoux) and #407 (JARPD) do track closely with each other,
indicating that the two locations are probably impacted by similar PM2.s sources (Figure 3.18). The two
locations are separated by residential areas, open space, and some light industrial complexes.
Rubidoux and JARPD, PM2 5 (|ig/m3)
405 407
80.0
O 70.0
u 50.0
C
o
U 40.0
c
2 30.0
5
a: 20.0
x
4 10.0
-------
Coachella Valley
The three units deployed in the Coachella Valley include CSAM #406 at Saul Martinez AMS, CSAM #409 at
Valley View Elementary School, and CSAM #410 at the Indio AMS. The deployment date for units #406 and
#409 was delayed until February 22, 2017 due to unfinished laboratory testing activities, and the Indio
AMS was deployed earlier, on January 17, 2017. The data recovery for CSAMs #406 and #409 for PM2.5 was
low, at 0.7% and 5.4%, respectively (the Alphasense OPC-N2 sensor readings were consistently below 1
Hg/m3). The time series for the Coachella Valley region is represented by the measurements provided by
the CSAM #410, deployed at the Indio AMS (Figure 3.19).
Coachella Valley
410
250
200
^ 150
OD
3
J 100
50
1/17/17 1/27/17 2/6/17 2/16/17 2/26/17 3/8/17
Figure 3.19. Time series of CSAM Pod #410 in Coachella Valley
A
i
Hjuu J
1
LmImIlL 1 jLmL 1 W
3/18/17 3/28/17 4/7/17
3.2.2. Ambient Ozone
The CSAM pods utilize an Aeroqual metal oxide ozone sensor. During the co-location testing at the Rubidoux
AMS, the ozone sensors correlated well with the reference instrumentation, with low intra- model
variability. As a result, differences among the sensors should be indicative of the quantitative differences
among the various locations. The Data recovery for the CSAM ozone sensor was high (>85%) during the field
deployment. Table 3.6 shows the summary statistics for ozone from the ten CSAM units.
Table 3.6. CSAM Ozone Summary Statistics
Ozone(ppb)
Long Beach
Jurupa Valley
Coachella Valley
401
402
404
403
405
407
408
406
409
410
Min
0.1
6.0
0.2
5.4
0.2
7.9
2.6
13.3
9.9
5.1
Max
80.3
36.0
45.9
50.8
63.2
50.1
82.5
38.8
36.9
57.3
Median
25.6
14.7
17.3
17.5
22.4
23.5
27.4
27.3
23.7
33.0
Mean
23.8
15.2
17.8
20.2
21.4
23.7
27.0
26.9
23.6
30.4
STDEV
14.16
4.32
9.08
7.69
14.36
8.19
9.98
4.78
5.80
10.82
SE
0.09
0.03
0.07
0.05
0.09
0.05
0.06
0.04
0.05
0.07
Recovery {%)
96.5
85.9
99.6
99.6
99.6
99.6
99.6
91.2
99.6
99.6
58
-------
The CSAM pods that were located at SCAQMD air monitoring stations equipped with ozone analyzers are the
pods at Hudson, Mira Loma, Rubidoux, and Indio. The summary statistics for the FEM ozone analyzers can be
found in Table 3.7.
Table 3.7. FEM Air Monitoring Station Ozone Summary Statistics
FEM Ozone (ppb)
Hudson
MLVB
Rubidoux
Indio
Min
0.0
0.2
0.0
0.1
Max
94.8
102.0
101.6
133.0
Median
25.5
26.5
27.9
40.9
Mean
23.8
27.3
27.6
38.8
STDEV
16.0
20.6
20.4
18.5
SE
0.10
0.13
0.12
0.11
Recovery{%)
91.7
93.8
94.5
98.5
The overall results are unsurprising, as they indicate a well-known diurnal variation of 03 in the air basin
related to the photochemical production of 03. The diurnal variability is characterized by high concentrations
during daytime hours and lower concentrations at nighttime and early in the morning. Throughout the study,
the mean concentration for the Long Beach sites was 18.9 ppb. The mean values in Jurupa Valley and
Coachella Valley were 23.1 and 27.0 ppb, respectively. Although the mean 03 concentration in the Coachella
Valley was higher than that measured in the Jurupa Valley, the maximum 5-minute average readings were
recorded at the Jurupa Valley site.
Long Beach
The lowest mean concentration for ozone among the Long Beach locations was at the near roadway site
along the 1-710. The 1-710 near-roadway location is considered an ozone sink, with localized emissions of NO
scavenging 03. The time series for the three Long Beach locations are shown in Figure 3.22. The highest
readings in this region were recorded by unit #401 at the Hudson AMS, with a mean concentration of 23.8
ppb. The summary statistics for unit #401 match well with the FEM ozone analyzer at Hudson AMS. When
comparing the 24-hour mean concentrations between CSAM 401 and the FEM at Hudson, the sensor tracked
the FEM ozone analyzer well, as shown in Figure 3.20.
Hudson FEM and CSAM 401
Hudson FEM 401
45
1/5/17 0:00 1/25/17 0:00 2/14/17 0:00 3/6/17 0:00 3/26/17 0:00 4/15/17 0:00 5/5/17 0:00
Figure 3.20. Hudson FEM and CSAM #401 time series
The scatterplot shown in Figure 3.21 between the FEM at Hudson and CSAM 401 indicates that the sensor
performed well throughout the field deployment stage. The R2 value was found to be 0.93 with a slope of
0.85.
59
-------
Hudson FEM and CSAM 401 _ Q g55x + 3 515g
24-Hr mean conc. r2 = 0.9301
•
* m,**
f
10 20 30 40
Hudson FEM ozone (ppb)
Figure 3.21. Hudson FEM and CSAM #401 scatterplot
For the entire duration of the study, the 8-hour ozone average level never exceeded the NAAQS (70 ppb) in
the Long Beach area, as shown in Figure 3.23.
Figure 3.22. Long Beach Ozone 24-hour time series
Figure 3.23. Long Beach Ozone 8-hour time series
60
-------
Jurupa Valley
The four CSAM pods (#403, #405, #407 and #408) deployed in the Jurupa Valley tracked well with one another
when considering their 24-hour averages (Figure 3.24). The unit deployed at the Mira Loma AMS recorded
the highest mean concentrations for the Jurupa Valley.
Jurupa Valley, Ozone (ppb)
403 405 407 408
Figure 3.24. Jurupa Valley Ozone Concentrations, 24-hour Mean
For the duration of the study, the 8-hour average ozone level in the Jurupa Valley area did not exceed the
NAAQS (70 ppb) as shown in Figure 3.25.
Jurupa Valley, Ozone (ppb)
403 405 407 408
60
1/12 1/17 1/22 1/27 2/1 2/6 2/11 2/16 2/21 2/26 3/3 3/8 3/13 3/18 3/23 3/28 4/2 4/7 4/12 4/17
Figure 3.25. Jurupa Valley Ozone Time series, 8-hour average
CSAM units #403 and #405 were co-located at the Rubidoux AMS. As the study progressed, the CSAM
#403 ozone sensor appeared to gradually lose sensitivity to changes in ozone concentrations. This gradual
trend was observed as the study progressed and became more apparent beginning around
February 26, 2017. Figure 3.26 displays the time series data of the two co-located ozone sensors and
the changes over time. When comparing the two sensors, the R2 value improved from 0.53 to 0.96, as
shown in Figure 3.27.a-b, when filtering the data prior to February 26, 2017. The co-location time series
and scatterplot for PM2.s (Figures 3.25 and 3.28) indicates a similar result, with an increase in R2 values
after filtering out data generated after February 26, 2017. The lessened performance of both the PM and
O3 sensors in CSAM #403 around the same time (February 26, 2017) indicates a potential systematic
problem (i.e., a power supply or board issue) rather than an individual sensor problem.
61
-------
Rubidoux Co-located, Ozone (ppb)
403 405
60
Figure 3.26. Rubidoux Co-Location Ozone time series, 8-hour average
a.
Rubidoux Co-located y = i.3569x - 6.0947
1-hravg R2 = 0.5281
70.0 ,
60.0
I 500 fr. A -
s3ao
lo 20.0
- mo
0.0 '
0.0 10.0 20.0 30.0 40.0 50.0 60.0
403 Ozone (ppb)
Figure 3.27. Rubidoux Co-Location Correlation Plots: a. Full time period, b. Filtered time to include
January 1, 2017 to February 26, 2017
The FRM ozone instrument at the Rubidoux station was found to correlate well with CSAM #405, with an
R2 = 0.94, and not to correlate well with CSAM #403 (R2=0.35), as shown in Figure 3.28.
0.00 10.00 20.00 30.00 40.00 50.00 0.00 10.00 20.00 30.00 40.00 50.00
FEM Ozone (ppb) FEM Ozone (ppb)
Figure 3.28. Rubidoux FRM ozone and CSAM #403 and #405 comparison
Coachella Valley
The Coachella Valley is in a desert region. Within the time frame of this study, the Indio AMS location
indicated the highest 24-hour mean concentrations for ozone among the three Coachella Valley locations.
The Coachella Valley's evening ozone readings rarely dropped below 10 ppb (Figure 3.29). This may be
because the land in the region is primarily devoted to agricultural use and does not have significant
62
-------
NO emissions from evening rush hour traffic that would titrate the ozone in the evening hours. During the
study period, the CSAM units did not report ozone concentrations that exceeded the NAAQS' 8-hour ozone
standard of 70 ppb (Figure 3.30).
Coachella, Ozone (ppb)
406 409 410
45.0
0.0
1/16 1/21 1/26 1/31 2/S 2/10 2/15 2/20 2/25 3/2 3/7 3/12 3/17 3/22 3/27 4/1 4/6 4/11 4/16
Figure 3.29. Coachella Valley Ozone, 24-hour Mean
Figure 3.30. Coachella Valley Ozone, 8-hour rolling average
The Indio AMS employs an FEM instrument to measure ozone. Upon a comparison between the Indio
AMS reference instrumentation and the CSAM #410, reduced sensor performance was observed over
time, as indicated in the time series (Figure 3.31). The scatterplot between the FRM at Indio and CSAM #410
is shown in Figure 3.32.
Indio FEM Ozone and CSAM 410
(RM 410
1/17 1/24 1/31 2/7 2/14 2/21 2/28 3/7 3/14 3/21 3/28 4/4 4/11 4/18
Figure 3.31. Indio FRM ozone and CSAM #410 time series
63
-------
Indio 03 FEM and CSAM 410 y = 0.5559X + 8.:
Overall: lhr mean R2 = 0.8807
70.0
— 60.0
_Q
Q.
3 50-°
a)
§ 40.0
M
O 30.0
O
5 20.0
10.0
0.0
0 20 40 60 80 100
FEM Ozone (ppb)
Figure 3.32. Indio AMS FRM and CSAM #410 scatterplot
The time series and scatterplot comparing CSAM #410 with the Indio AMS FRM for ozone indicates that
the sensor lost sensitivity to ozone over time. Table 3.8 shows the changes in slope, intercept, and R2 value
between the FRM and CSAM #410 ozone sensor by examining the deployment month by month. Over four
months, the slope between the reference and the CSAM was reduced from 0.84 in January to 0.47 in April,
whereas the intercept increased from 4.1 to 10.1 over the same time period. The R2 stayed consistent over
the four months and was greater than 0.95.
Table 3.8. FRM and CSAM #410 ozone sensor regression parameters by month
Time period
Slope
Intercept
R2
Overall
0.56
8.8
0.88
January*
0.84
4.1
0.98
February
0.69
5.4
0.97
March
0.57
7.7
0.95
April*
0.47
10.1
0.97
*CSAM not dep
oyed entire month
When using the Indio AMS as a reference for the Coachella Valley region, it may appear that all three of the
CSAM ozone sensors were losing sensitivity to ozone overtime in an examination of the FRM ozone readings
with the three monitors on a 24-hour average (Figure 3.33). However, such losses in responsiveness over
time are to be expected with this type of sensor. CSAM units #406 and #409, which underwent laboratory
testing at relatively high challenge concentrations, might have also experienced an impact on their lifespans.
An underestimation of ozone concentrations in the region by low-cost sensors deployed in a community-
based monitoring program will not provide accurate data without correction factors. The underestimation
may miss potential days in which the NAAQS would have been exceeded if the sensor had been performing
correctly.
64
-------
Indo FEM and Coachella Valley CSAM
FRM 406 409 410
70.0
0.0 I 1 1 1 1 1 1 1 1 1
1/16/17 1/26/17 2/5/17 2/15/17 2/25/17 3/7/17 3/17/17 3/27/17 4/6/17 4/16/17
Figure 3.33. Time series of FRM and CSAM Pods #406, #409 and #410 (24-hour average)
3.2.3. Temperature and RH influences
Environmental factors, such as temperature and RH, may affect the measurement of air pollutants by low-
cost sensing technology. Optical particle counters that measure scattered light from particles to count, size,
and calculate a particulate mass value are susceptible to moisture affects. Particles containing high RH or
moisture affect the light- scattering properties of measured aerosol; consequently, costly optical particulate
counters typically include a dryer or heater to maintain the RH within a desired range for measurement. An
estimation for dew point (DP), which includes both temperature and RH, provides the opportunity to
determine if there is a combined effect of temperature and RH on the sensor's response. An approximation
for the DP temperature suggested by Lawrence (2005) is useful for moist air (RH > 50%) and valuable for
examining the dew point temperature with respect to the OPC measurement response, as the concern for
measurement influences with RH < 50% is low.
/100 — RH\
Dv~l~ ( 5 )
The measurement of PM25 and ozone are plotted against temperature, RH, and dew point. Temperature,
RH, and dew point do not affect the measurement of ozone. The following three plots represent the Indio
CSAM #410's ozone response with the three environmental parameters. These plots are characteristic of
the ten CSAM pods for ozone. Figure 3.34 indicates ozone response to temperature: as temperature rises,
the ozone values rise as well. This trend is expected, as sunlight is necessary for the production of ozone. As
humidity decreases, the ozone values increase, as shown in Figure 3.35. This trend also is expected, as ozone
concentrations peak during the early afternoon hours, which is also a time period characteristic of low
humidity. No trend or correlation is evident between ozone and the approximation for dew point, as shown
in Figure 3.36.
65
-------
60
50
40
30
Indio 410
Temperature and Ozone
Indio 410
Humidity and Ozone
§@i
20 30
Temp (°C)
Figure 3.34. Iridio 410 Temperature vs. Ozone
Relative Humidity (%)
Figure 3.35. Indio 410 Humidity vs. Ozone
Indio 410
Dew point and Ozone
J*!'
V
£%• • •
5 10 15 20 25
Dew Point Temperature (°C)
Figure 3.36. Indio 410 Dew Point and Ozone
The PM2.s mass concentrations of the OPC sensor were plotted against the three environmental factors of
temperature, RH, and dew point (Figures 3.37-3,45). Plots for one sensor from each of the three major
geographical areas (Long Beach, Jurupa Valley, and Coacheila Valley) are shown below. The plot for CSAM
#401 at Hudson AMS in the Long Beach region indicates that high PM concentrations are associated with
high RH values. When reviewing the dew point, which includes both temperature and RH, a more refined
peak than the T/RH plots is observed, with a peak centered in the 11-13 °C range for high PM concentrations.
Hudson 401 Temperature and PM2
Hudson 401 Humidity and PM2
450
400
£
350
oo
3
300
~~Z>
250
2
200
CL
>•
150
3
o
100
X
50
•
•
I
s «•
• •
V;
JmW
•
«
10 15 20 25 30 35 40
Temp (°C)
450
400
g 350
M 300
~ 250
2E 200
Q.
>• 150
0 100
1
50
0
20
' • V
• l:£
v * &%
. !«r''
""iS
40 60
Relative Humidity (%)
Figure 3.37 Hudson 401 Temp and PM
Figure 3.38. Hudson 401 Humidity and PM
66
-------
Figure 3.39. Hudson 401 Dew Point and PM
The plot for CSAM #407 in the Jurupa Valley region indicates that high PM concentrations are associated
with high RH values and low temperatures. The comparison between temperature and dew point in the
Jurupa Valley indicates more scatter than the Long Beach sites. This may be because the inland valleys
experience more extreme weather conditions than the ocean communities, where conditions are tempered
by the ocean's influence.
JARPD 407 Temperature and PM2 5
350
tr 300 • *
E
-g 25°
200 # # # •
| 150 | |^| 1
f 100 | • u • .•
X 50 •
0 I—¦
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46
Temp (°C)
Figure 3.40. JARPD 407 Temp and PM
JARPD 407 Dew point and PM2 5
350
300
•
250
•
• •
§
• •
•
•
•
•»
•
• 1
ff__ *3
•
j£\
JARPD 407 Humidity and PM2 5
40 60
Relative Humidity {%)
Figure 3.41. JARPD 407 Humidity and PM
5 10 15
Dew Point Temperature (°C)
Figure 3.42. JARPD Dew Point and PM
The plot for CSAM #410 at the Indio AMS in the Coachella Valley region indicates that high PM
concentrations are associated with high RH values. The high RH events coincided with temperatures around
11.5 °C, which also indicated high PM values. As a result, Figure 4.45 represents the dew point in comparison
to the PM concentrations and indicates a resolved peak around 11.5 °C. This is similar to the values observed
for the Long Beach area.
67
-------
200
180
160
140
120
100
80
60
40
20
0
Indio 410
Temperature and PM2 5
J.
Temp (°C)
Figure 3.43. Indio 410 Temp and PM
200
E
180
160
3.
140
120
100
O-
>•
80
3
60
O
X
40
20
0
Indio 410
Humidity and PM2 5
20 40 60
Relative Humidity (%)
Figure 3.44. Indio 410 Humidity and PM
Figure 3.45. Indio 410 Dew Point and PM
In summary, no correlation between the temperature, RH, or dew point was found for the ozone or PM
sensors. Additional research could further investigate the effects and potentially apply filters for specific
temperature and RH values that may lead to invalid measurements by the OPC.
3.3. QA summary for field deployment
Throughout the CSAM field deployment, the Atmospheric Measurements (AM) Branch at SCAQMD operated
and maintained the regulatory air monitoring stations across the air basin. Seven of the CSAM pods were
located at six regulatory air monitoring stations. The ozone and particle instrumentation at these monitoring
stations were maintained and operated in accordance with standard air monitoring practices. The FEM
ozone instruments were maintained and checked weekly by station operators to verify that the instrument
was operating according to the specifications defined in the SCAQMD's SOPs. The ozone instruments were
subjected to daily precision and weekly span checks to verify the instrument was operating within 7% of the
expected values for precision and span checks. During the field deployment, the ozone instruments at the
Indio AMS and Rubidoux AMS were all within 7% of the expected values for all precision and span checks.
These instruments were calibrated every six months or as needed if found to be out of tolerance. The ozone
instrument at Indio was calibrated on December 10, 2016 and the instrument at Rubidoux was calibrated
on November 2, 2016. The quality assurance branch at SCAQMD performs an ozone concentration ramping
audit on the ozone analyzers annually. The combined maintenance, calibration schedule, and audit
schedules performed on the station ozone instrumentation provides assurance that the data are of high
quality.
The equivalent method particle instrumentation at the stations were subjected to flow, leak, temperature,
68
-------
and pressure verifications monthly. Throughout the course of the field deployment, the flow, leak,
temperature, and pressure verifications performed on the BAM units were found to be within specifications.
The Metone BAM also automatically performs a span check each hour with a reference membrane to ensure
that the instrument is not drifting over time. The Metone BAM units are calibrated every six months and the
quality assurance branch performs audits of the BAM twice a year. The MLVB BAM was calibrated on
December 6, 2016 and the Rubidoux BAM was calibrated on June 24, 2016 and February 10, 2017. The
Grimm 180EDM undergoes flow checks monthly and is calibrated annually. The most recent calibration of
the Grimm 180EDM was performed by the factory on November 29, 2016. The combined maintenance,
verifications, and checks on the FEM equipment provides assurance that the instrumentation was operating
in good condition and producing quality data.
The CSAM units were checked once every two weeks to ensure proper data logging and to verify date and
timestamps. No further quality assurance checks or verifications were performed during the field
deployment.
69
-------
Conclusions
The AQ-SPEC group successfully evaluated the performance of EPA-designed CSAM pods both in the field
and laboratory. The pods were then successfully deployed at nine locations across southern California in
three distinct areas including Long Beach, Jurupa Valley, and Coachella Valley. The extended deployment
and subsequent data analysis revealed some of the concerns related to long-term sensor deployment,
including measurement sensitivity change and sensor degradation over time. The AQ-SPEC group
recommends re-engineering the pods to include built-in data communication. The ability to remotely access
the data would allow data validation and analysis procedures to be built into cloud-based applications that
could detect sensor performance losses and failures on a timelier basis. This would reduce the need for staff
members to visit the sites to download and compile the data, and then to perform statistical analysis weeks
after the measurements have been taken. In addition, the AQ-SPEC recommends performing extended
sensor performance testing before entering into sensor selection and subsequent sensor development for
ambient air quality monitoring. Care should be taken to identify the raw sensor with the best fit within the
allocated budget that provides an adequate sensor life time for the duration of the study. Minimizing down
time to reduce data losses is also important when selecting sensors for in-field deployment. The AQ-SPEC
also recommends that sensors that have undergone extensive field and laboratory testing should not be
used in subsequent field deployments. Extensive testing in areas characterized by elevated air pollution
levels may damage or alter the sensors' performance, which could result in significant data loss after the
sensors have been deployed in the field for specific air monitoring applications.
70
-------
References
Barzyk, T., R. Williams, A. Kaufman, M. Greenberg, M. OShea, P. Sheridan, A. Hoang, A. Teitz, C. Ash, M.
Mustafa, AND S. Garvey. 2016. Citizen Science Air Monitoring in the Ironbound Community. U.S.
Environmental Protection Agency, Washington, D.C., EPA/600/R-16/049.
https://cfpub.epa.gov/si/si public record report.cfm?dirEntryld=315154, last accessed September 18,
2017.
Jiao, W., Hagler, G., Williams, R., Sharpe, B., Brown, R., Garver, D., Judge, R., Caudill, M., Richard, J., Davis,
M., Weinstock, L., Zimmer-Dauphinee, S., Buckley, K. 2016. Community air sensor network (CAIRSENSE)
project: Evaluation of low-cost sensor performance in a suburban environment in the southeastern United
States. Atmospheric Measurement Technology, DOI:10.5194/amt-9-5281-2016.
Jiao, W., Hagler, G., Williams, R., Sharpe, B., Brown, R., Garver, D., Judge, R., Caudill, M., Richard, J., Davis,
M., Weinstock, L., Zimmer-Dauphinee, S., Buckley, K. "Community Air Sensor Network (CAIRSENSE) Project:
Lower Cost, Continuous Ambient Monitoring Methods", presented at the 108th Annual Meeting of the Air &
Waste Management Association, June 22-25, 2015 in Raleigh, NC.
https://www.epa.gov/sites/production/files/2015-07/documents/cairsense.pdf, last accessed September
18, 2017.
Lawrence, M.G. 2005. The relationship between relative humidity and the dewpoint temperature in moist
air: A simple conversion and applications, American Meteorological Society, DOI:10.1175/BAMS-86-2-225.
Papapostolou, V., Zhang, H., Feenstra, B.J., and Polidori A. 2017. Development of an environmental chamber
for evaluating the performance of low-cost air quality sensors under controlled conditions. Atmospheric
Environment, 171: 82-90.
Polidori, A., Papapostolou, V., Zhang, H. 2016. Laboratory Evaluation of Low-Cost Air Quality Sensors -
Laboratory Setup and Testing Protocol. South Coast Air Quality Management District, Diamond Bar, CA.
South Coast Air Quality Management District Final 2016 Air Quality Management Plan (2016 SCAQMD
AQMP). http://www.aqmd.gov/home/librarv/clean-air-plans/air-quality-mgt-plan/final-2016-aqmp, last
accessed September 18, 2017.
Williams, R., A. Kaufman, and S. Garvey. 2015a. Next Generation Air Monitoring (NGAM) VOC Sensor
Evaluation Report. U.S. Environmental Protection Agency, Washington, DC, EPA/600/R-15/122 (NTIS
PB2015-105133). https://cfpub.epa.gov/si/si public record report.cfm?dirEntrvld=308114, last accessed
September 18, 2017.
Williams, R., A. Kaufman, T. Hanley, J. Rice, and S. Garvey. 2015b. Evaluation of Elm and Speck Sensors. U.S.
Environmental Protection Agency, Washington, DC, EPA/600/R-15/314.
https://cfpub.epa.gov/si/si public record report.cfm?dirEntryld=310285, last accessed September 18,
2017.
Williams, R., T. Barzyk, and A. Kaufman. 2015c. Citizen Science Air Monitor (CSAM) Operating Procedures.
U.S. Environmental Protection Agency, Washington, DC, EPA/600/R-15/051.
https://cfpub.epa.gov/si/si public record report.cfm?dirEntryld=307111&simpleSearch=l&searchAII=CS
AM+Operating+Procedures, last accessed September 18, 2017.
71
-------
Williams R., A. Kaufman, T. Hanley, J. Rice, and S. Garvey. 2014a. Evaluation of Field-deployed Low Cost PM
Sensors. U.S. Environmental Protection Agency, Washington, DC, EPA/600/R-14/464 (NTIS PB 2015-102104).
https://cfpub.epa.gov/si/si public record report.cfm?dirEntryld=297517, last accessed September 18,
2017.
Williams, R., R. Long, M. Beaver, A. Kaufman, F. Zeiger, M. Heimbinder, I. Hang, R. Yap, B. Acharya, B.
Ginwald, K. Kupcho, S. Robinson, O. Zaouak, B. Aubert, M. Hannigan, R. Piedrahita, N. Masson, B. Moran, M.
Rook, P. Heppner, C. Cogar, N. Nikzad, and W. Griswold. 2014b. Sensor Evaluation Report. U.S.
Environmental Protection Agency, Washington, D.C., EPA/600/R-14/143 (NTIS PB2015-100611),
https://cfpub.epa.gov/si/si public record report.cfm?dirEntryld=277270, last accessed September 18,
2017.
U.S. Environmental Protection Agency (EPA). 2016. Village Green and AirMapper Fact Sheet.
https://www.epa.gov/air-research/village-green-and-airmapper-fact-sheet, last accessed September 18,
2017.
72
------- |