oEPA
United States
Environmenta
Agency
LPA/600/R-17/171 I July 2017 I vwvw.epa.gov/research
Performance Evaluation of
the United Nations
Environment Programme
Air Quality Monitoring Unit
Office of Research arid Development
National Exposure Research Laboratory

-------
EPA 600/R-17/171
July 2017
Performance Evaluation of the United
Nations Environment Programme Air
Quality Monitoring Unit
Ron Williams, Teri Conner, Andrea Clements
National Exposure Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Research Triangle Park, NC, USA 27711
Valentin Foltescu, Victor Nthusi, Jason Jabbour
United Nations Environment Programme
Science Division
Washington, DC 20006
David Nash, Joann Rice, Amanda Kaufman
Office of Air Quality Planning & Standards
U.S. Environmental Protection Agency
Research Triangle Park, NC, USA 27711
Alexis Rourk
Office of International and Tribal Affairs
U.S. Environmental Protection Agency
Washington, DC 20004
Manu Srivastava
Jacobs Technology Inc.
Research Triangle Park, NC, USA 27711

-------
Disclaimer
This technical report presents the results of work performed by the US. Environmental Protection
Agency's Office of Research and Development (ORD) with technical support through Jacobs
Technology (Contract # EP-C-15-008) for the National Exposure Research Laboratory (NERL),
U.S. Environmental Protection Agency (US EPA), Research Triangle Park, NC. The effort
represents a collaboration between the US EPA and the United Nations Environment Programme
(UNEP) in fulfillment of a Material Cooperative Research and Development Agreement
(MCRADA) to conduct research involving emerging air quality sensor technology. It has been
reviewed by the U.S. EPA and the UNEP and approved for publication. Mention of trade names
or commercial products does not constitute endorsement or recommendation for use.
1

-------
Table of Contents
Disclaimer	
Table of Contents	
List of Figures	i
List of Tables	h
Acronyms and Abbreviations	
Acknowledgments	\
Executive Summary	v
1.0 Introduction	
2.0 Materials and Methods	
2.1	Instrumentation	
2.2	Deployment	'
2.3	Data Collection Procedure	
2.4	Data Processing	
2.5	Data Processing Observations	1
2.6	US EPA Quality Assurance Review and Application	1
3.0 UNEP Pod Results and Discussion	1
3.1	RH Comparison	1
3.2	Temperature Comparison	1
3.3	PM Mass Concentration Comparisons	1
4.0 Ease of Use Features and Concerns	2
4.1	Hardware	2
4.2	Data Processing	2
4.3	Study Limitations	2
4.4	Conclusions	2'
5.0 References	2
6.0 Appendix	2
li

-------
List of Figures
Figure 1. GRIMM monitor (left) and select reference monitors at the AIRS	6
Figure 2. Orientation of UNEP pod and associated wiring connections	7
Figure 3. The aluminum test shelter with the reference monitor in the background	7
Figure 4. Select AIRS reference monitor inlets	8
Figure 5. Example of processed file	9
Figure 5a. Example of processed file (continued)	10
Figure 6. Matched time stamp data for UNEP Pod and AIRS reference data	12
Figure 7. Linear Regression of the UNEP pod versus the AIRS reference RH	14
Figure 8. Time series showing the response offset of the UNEP pod versus the AIRS
reference RH	14
Figure 9. PM response relative to ambient RH conditions	15
Figure 10. Regression of 24-hr average PM concentration from the UNEP pod versus
reference RH	16
Figure 11. Regression of 1-hr average PM2.5 concentration from the UNEP pod versus
reference RH	17
Figure 12. Regression of 1-hr average PM10 concentration from the UNEP pod versus
reference RH	17
Figure 13. Time series comparison of UNEP pod versus AIRS reference temperature	18
Figure 14. Regression of UNEP pod versus AIRS reference temperature	18
Figure 15. Regression of 24-hr average UNEP pod PM response versus reference
temperature	19
Figure 16. 24-hr average PM2.5 concentration time series for the UNEP pod and AIRS
reference monitor	20
Figure 17. 24-hr average PM10 concentration time series for the UNEP pod and AIRS
reference monitor	20
Figure 18. UNEP pod versus reference 24-hr average PM2.5 concentration	21
Figure 19. UNEP pod versus reference 24-hr average PM10 concentration	21
Figure 20. 5 minute UN EP pod response versus ambient RH at study onset	22
Figure 21. 1-hour average UNEP pod versus AIRS reference PM2.5 concentration	23
Figure 22. 1-hour average UNEP pod versus AIRS reference PM10 concentration	23
Figure A1. PuTTY command code	29
111

-------
Figure A2. PuTTY command code (continued)	30
Figure A3. PuTTY command code (continued)	30
Figure A4. PuTTY command code (continued)	31
Figure A5. PuTTY command code (continued)	31
Figure A6. Example file directory	32
Figure A7. Example hourly data file directory	33
Figure A8. Example of suspected monthly data file reporting error	34
Figure A9. Example of suspected monthly data file reporting error (continued)	34
Figure A10. Anomalous concentration spikes	35
Figure A11. Anomalous concentration spikes (continued)	35
Figure A12. Repetitious time stamps	35
List of Tables
Table 1. Impact of applying > 95% RH exclusion criteria on the strength of the UNEP
pod versus reference PM mass concentration comparison (reported as R2 values) at
various averaging times	16
Table 2. Impact of averaging time on regression results comparing UNEP pod versus
reference PM mass concentration	23
iv

-------
Acronyms and Abbreviations
AC
alternating current
AIRS
Ambient Air Innovation Research Site
FEM
federal equivalent method
FRM
federal reference method
GPS
Global Positioning System
MCRADA
Material Cooperative Research and Development Agreement
NAAQS
National Ambient Air Quality Standards
NERL
National Exposure Research Laboratory
N02
nitrogen dioxide
OAQPS
Office of Air Quality Planning and Standards
OITA
Office of International and Tribal Activities
ORD
Office of Research and Development
PM
particulate matter
PM2.5
particulate matter of diameter 2.5 microns or less
PM10
particulate matter of diameter 10 microns or less
ppb
parts per billion
ppm
parts per million
QAPP
quality assurance project plan
R2
coefficient of determination
RH
relative humidity
ROP
research operating procedure
RTP
Research Triangle Park
S02
sulfur dioxide
UNEP
United Nations Environment Programme
US EPA
United States Environmental Protection Agency
VOC
volatile organic compound
V

-------
Acknowledgments
The NERL's Quality Assurance Manager (Sania Tong-Argao) is acknowledged for her
contributions to the development of research operating procedures (ROPs) and the project's
quality assurance project plan (QAPP) used in execution of the research effort. Sarah Bauer (US
EPA's Federal Technology Transfer Act Program) is acknowledged for her efforts involving
establishment of the MCRADA between the US EPA and the UNEP.
vi

-------
Executive Summary
A request for technical collaboration between the UNEP and the US EPA resulted in the
establishment of a MCRADA. The purpose of this agreement was to evaluate a prototype air
quality monitoring system (referred to as the UNEP pod) developed by the UNEP for use in
environmental situations where more sophisticated monitoring instrumentation was not available.
The US EPA has conducted numerous evaluations of other similar sensor pods at its Research
Triangle Park, NC research campus and has trained staff as well as established research designs
for such efforts. Under the terms of the MCRADA, the US EPA would operate the pod using
UNEP-provided operating procedures in a manner consistent with its planned intent of
deployment. The US EPA would collect air quality monitoring data from the pod for selected
environmental measures over a period of approximately 1 month. Reference monitoring data
collected from collocated federal regulatory monitors would be used to establish a comparison
between the two systems and thus establishment of performance characteristics. In addition, the
US EPA would provide informed feedback to the UNEP about the pod's observed ease of use
features that would be beneficial in its future evolution and deployment.
Study Objectives
In response to the UNEP's request, the US EPA evaluated the sensor pod during a 30-day study
to establish its basic performance characteristics. The effort was projected to be initiated in the
fall of 2016 and fully completed during calendar year 2017. Specifically, the US EPA agreed to
the following:
•	Conduct a collocated comparison between the UNEP low cost sensor pod versus
reference monitors under outdoor environmental conditions at the US EPA's research site
in Research Triangle Park, NC for a period of approximately 1 month.
•	Collaborate with the UNEP on all data summaries.
•	Publish basic summary findings of the effort following peer review in a mutually agreed
upon format (e.g., peer reviewed report).
•	Provide a complete database detailing both the UNEP sensor pod and reference monitor
response under collocated ambient challenge conditions to allow the UNEP to conduct its
own independent statistical comparison of performance.
Study Approach
The UNEP sensor pod was operated under UNEP operating guidelines on an "as is" basis. That
is, the US EPA technically sited and operated the prototype unit as defined by the UNEP to
ensure that the performance characteristics established were representative of those to be
expected from real world deployment. Although the US EPA requested that multiple copies of
the device be provided to allow its precision, bias, and other performance characteristics to be
established, only a single pod was made available. Whereas evaluation of the pod under fully
controlled laboratory conditions would have been valuable (i.e., a chamber), no such chamber
was immediately available to the US EPA for such an effort nor were there resources to conduct
such an effort. Therefore, the pod was operated in a weatherproof enclosure in a collocated
manner at the US EPA's Ambient Air Innovation Research Site (AIRS) for a period of
vii

-------
approximately 1 month. One (1) minute continuous data responses (PM2.5, PM10, NO2, SO2, RH
and temperature) were obtained. Data were continuously logged to the internal microprocessor
and downloaded to a dedicated laptop computer weekly. Using applications provided by the
UNEP, data were combined into a time series, converted to ambient concentrations, and recorded
in electronic spreadsheets. US EPA validation of the data was performed following consultation
with the UNEP on all issues that required input on data validation resolution. Reference
monitoring data from the AIRS were obtained and, following validation review, integrated with
UNEP pod data to allow for statistical evaluation and characterization. The statistical comparison
of the UNEP pod versus the reference monitor data was compared to established performance
characteristics (e.g., time series, regression, co-linearity with RH and temperature). Statistical
averaging times ranging from 1 minute to 24 hours were explored to establish integration effects
on performance characteristics. The impact of various quality assurance data inclusion/exclusion
criteria were considered. Statistical findings, based on validated data meeting specific quality
assurance criteria that were established, are presented in this report.
Sensor Performance Results
The UNEP pod was determined to have a stable electronic data collection architecture
(microprocessor and integrated sensor components) in that data were collected without failure
from the initiation of the study through its completion (for more than 30 days). However,
processing errors using the UNEP developed firmware and software that complicated data
analysis and quality assurance review were noted. Even so, a data completeness record of > 90%
was observed.
Preliminary observations of raw (non-validated) NO2 and SO2 data following the first week of
data collections indicated that the UNEP pod's response was significantly different in terms of its
reported concentrations than those measured by the collocated reference monitors. Consultation
with the UNEP on these observations resulted in a decision to continue data collection for these
pollutants, but not to pursue establishment of their performance characteristics as part of this
report. Raw data concerning these pollutants were harvested for the full duration of the
evaluation and have been shared with the UNEP for their consideration.
PM2.5, PM10, RH and temperature data from the UNEP pod were compared with the collocated
reference or research grade monitors. Briefly, time series, as well as regression analyses,
revealed little to no agreement (R2 < 0.1) of the PM2.5 and PM10 mass concentrations with
reference values at time averaging intervals between 5 minutes to 24 hours. Ambient RH was
determined not to be correlated to pod PM response over any integration period. Direct
comparison with collocated reference temperature data indicated that the UNEP pod's internal
components heated the airspace within the UNEP pod, resulting in reduced RH and a ~ 7°C
increase in temperature as measured by the UNEP pod. The impact, if any, that this heating
might have had on the resulting particulate matter or gas measurements is unknown.
Ease of Use Features Evaluation
The UNEP pod was observed to be very robust with respect to its day-to-day operation. It was of
sturdy construction relative to its physical design. However, even with the technical manual
provided by the UNEP, we were unable to establish some of the system's primary features (e.g.,
inlets, exhausts). An improved operator's manual would be beneficial to others. The US EPA did

-------
not open the pod to reveal its inner components (sensors and related electronics) to maintain the
integrity of its character. Once initialized using a standalone laptop computer, the UNEP pod
operated without electronic failure for more than 1 month of continuous 1-minute data
collections. Direct land power (alternating current) was used to provide the needed energy
resources. Data were logged to the internal microprocessor continuously. Data were downloaded
to a dedicated laptop computer weekly to ensure the UNEP pod was operating and storing data.
Accessing these data required some degree of technical capability. In particular, multiple third-
party software applications that were needed to provide the interface between the UNEP pod's
microprocessor and the laptop computer had to be downloaded from internet sources. We are
concerned that other end users, especially those with limited technical skills, might have some
difficulty not only in obtaining the secondary computer applications but also in their operation.
Two independent executables were used to process the raw data. These executables were
provided by the UNEP, and the US EPA attempted to use them in their "as is" state. The first, a
Windows command tool, was used to combine the 24 individual hourly data files into a single
combined data file representing a full day of data. The second, an Excel macro, converted the
raw gas data (reported as volts) to environmentally relevant concentration units (ppb). Some
coding errors were discovered in the macro code used to compute gas concentrations. The code
was revised, and all raw data were reprocessed using the updated code. The processing was
therefore not without some concerted effort to ensure completeness. Nevertheless, the
executables provided by the UNEP were robust. Since the UNEP developed the executables for
the US EPA to eliminate cloud-based processing (which the US EPA does not have permission
to use), it is unknown if the code script error was isolated to this specific situation or a more
systematic issue with primary data processing that might be encountered by others using the data
streaming capabilities of the UNEP pod.
Conclusions
The UNEP pod collected environmental data continuously with little need for operator
interaction. While in the US EPA's possession, the UNEP pod appeared to operate without
technical failure of any of the energized components (fans, microprocessor, sensor components).
Nevertheless, significant technical effort was required to obtain and operate the secondary
software applications that were needed to collect and download the data. The gas phase sensors
did not provide data useful for the evaluation. It is not clear if the problems with the gas sensors
were related to the algorithms used to process the data, or to the sensors themselves. Technical
suggestions were provided to the UNEP on this topic, which they may wish to pursue. Details are
presented in the Appendix.
While the UNEP pod successfully collected and reported data for both PM2.5 and PM10 mass
concentrations, comparisons over a wide range of time intervals failed to yield satisfactory
agreement with collocated reference monitors (R2 <0.18). Factors that might have had a negative
impact on the assessment (data completeness, RH impact) were exhaustively investigated, but
little to no improvement in the observed performance characterization was observed regardless
of data treatment.
IX

-------
1.0 Introduction
The UNEP recently developed a prototype multipollutant sensor pod called the UNEP Air
Quality Monitoring Unit, herein simply defined as the UNEP pod (http://aqicn.org/faq/2015-10-
28/unep-air-qualitv-monitoring-station/Y First introduced in 2015, the UNEP pod was developed
with the goal of providing an affordable air quality monitoring instrument to a worldwide
audience. A basic cost of - $1500/pod has been reported, which would potentially allow many
end users to obtain the device and operate it to achieve environmental monitoring for a variety of
needs. Basic features of the UNEP pod include a weatherproof milled aluminum encasement, a
series of gas phase sensors for criteria pollutant monitoring, an optical particle counter for
estimations of particulate matter mass, and environmental meteorological sensors, as well as fans
and other assorted electronic components. By design, the unit was intended for near-continuous
operation when energized with local land power, with data transmission occurring via cellular
communication to a dedicated service provider.
Although the UNEP pod was developed in 2015 and has undergone informal operational trials
since its release, no formal evaluation or reporting of its performance characteristics had been
conducted previously. Since emerging sensor technologies are not certified for their capabilities,
and the goal of the UNEP was to establish a credible air monitoring system for less developed
international settings, conducting such an evaluation was a defined need.
The US EPA has a documented history of conducting numerous technical performance
evaluations of low cost (< $2500) sensors and their assembled components (EPA, 2017a). Other
air quality research organizations have also begun to investigate sensor performance
characteristics (EPA, 2017b; EPA, 2017c). Under such scenarios, sensor components are tested
either directly under chamber (laboratory) challenge or under ambient (field) scenarios
(Williams, 2014b; Williams, 2014c; Williams, 2015a; Williams, 2015b). In all such cases, sensor
response has been directly compared versus Federal Reference Method (FRM) or Federal
Equivalent Method (FEM) monitors to establish the performance characteristics, and this
approach has been the subject of scientific discussion (Kaufman 2014; Williams 2014a; Williams
2014d). The general consensus has been that such an approach provides a peer-acceptable
method of defining the basic performance characteristic of a non-regulatory air quality device.
Specifically, the direct comparison of such a device under ambient (real-world) monitoring
conditions not only challenges the sensor to the pollutant(s) of interest but also to potential co-
factors to which the device might respond.
In the late winter of 2016, meetings were held involving representatives of the UNEP and the US
EPA's Office of International and Tribal Activities, ORD, and Office of Air Quality Planning
and Standards to discuss the possibility of conducting a formal performance review of the
UNEP's sensor pod. Ultimately, these meetings resulted in a formal MCRADA being established
between the UNEP and the US EPA's ORD to accept receipt of one or more of the sensor pods
and to conduct a month-long evaluation of its performance under real-world (ambient)
conditions. This research was to be conducted at the US EPA's AIRS test platform located on its
campus in Research Triangle Park, NC. This location was chosen solely because of its
convenience for the ORD staff who would be conducting the research. As this research was
being conducted without any additional funding or resources being made available, economics
1

-------
dictated that leveraging of existing resources (e.g., fully operational reference monitoring
platform and local US EPA staff) be used to achieve the primary goals. Both parties recognized
that the AIRS test location did not represent the environmental conditions that might be expected
in international locations where environmental pollution levels would be expected to be
significantly higher. Even so, ambient pollution levels of particulate matter as well as of the
National Ambient Air Quality Standards (NAAQS) criteria pollutant gases are routinely
measured above detection limits at the AIRS test location and with sufficient day-to-day
variability to enable an evaluation to be performed. The US EPA has previously reported on
using the AIRS test location in this manner to evaluate a wide range of lower cost sensor
components and multipollutant pod systems (Williams 2014b; Williams 2014d; Williams 2015a;
Williams 2015b).
The MCRADA established that the performance characterization research would be
collaborative in nature, with both parties substantially contributing to achieve its success.
Hallmarks of these contributions would be the UNEP providing support documentation fully
defining the operational guidelines of the UNEP pod and the US EPA operating the equipment as
expressly defined by the UNEP to ensure a non-biased approach to the testing. Other key
components of the MCRADA were as follows:
•	Operate the UNEP pods through a ~30-day study to establish its basic performance
characteristics, with a projected initiation in the fall of 2016 and completion during
calendar year 2017.
•	Conduct a collocated comparison between the UNEP pod versus the reference monitors
under outdoor environmental conditions at the US EPA's research site in Research
Triangle Park, NC. Ideally, the US EPA would test three UNEP pods side-by-side to see
if the results were replicable.
•	Collaborate with the UNEP on all data summaries.
•	Publish basic summary findings of the effort, following peer review, in a mutually agreed
upon format (e.g., peer-reviewed report).
•	Provide a complete database detailing both the UNEP pod and reference monitor
response under collocated ambient conditions to allow the UNEP to conduct its own
independent statistical comparison of performance.
Efforts were made to obtain the UNEP pods as early as possible in the calendar year, as the US
EPA had previously established, in laboratory testing, that gas phase sensor evaluations
performed under low temperature conditions (at or below freezing) sometimes resulted in poor
sensor performance (Williams, 2014d). It was the US EPA's goal to examine the sensor under
the most favorable conditions possible, and ultimately an agreement was reached to deliver the
pods in October 2016. Although a sustained effort was made to obtain multiple copies of the
UNEP pod to enable precision and bias evaluation, only a single pod was made available for the
research. The unit was recovered from a Kenya, Africa deployment and shipped to the US EPA's
Research Triangle Park campus, where the materials were unpacked, inspected, and cataloged. A
single screw was observed to be unattached in the packaging materials, and it was assumed that
the screw was originally secured inside the pod encasement itself and had become free because
of some movement or vibration during shipping and handling. Consultation with the UNEP on
2

-------
this matter yielded the same conclusion as US EPA staff that this item would not be expected to
have any impact on the resulting performance evaluation.
The software modules required to communicate with the pod and download the data, all
available as free shareware, were downloaded to a dedicated laptop. This software included: 1)
BONE_D64.exe ver. 1.2.0.715 - a BeagleBone serial over USB driver, 2) PuTTY.exe ver.
0.67.0.0 terminal emulator software, and 3) psftp.exe ver. 0.67.0.0, a secure file transfer protocol
for PuTTY. It is unknown whether these applications would have been required if the device was
being used in its cellular data transmission mode, as is the case at UNEP's Kenya-based site. The
dedicated laptop computer was used to communicate with the pod, to establish its operating
parameters, and ultimately to recover and process the raw data on a weekly basis. Two
executable files (Merge_CSV.cmd and Gas_ppb.xlsm), which were provided to the US EPA by
the UNEP, were needed to process the raw data. The first of these executables, a DOS command
file (Merge_CSV.cmd), merged the hourly data files into a single daily file containing all
measured data with raw gas data reported in units of volts. The second executable, an Excel
macro script (Gas_ppb.xlsm), transformed the raw gas data into environmentally relevant
concentration units (ppb). The result was a single Excel file for each day of data collection
containing the processed gas concentrations, particulate matter concentrations, temperature, and
RH, with each data record within the file identified by date and time. Data from these processed
data files were combined and used by the US EPA in its comparison with the collocated
reference monitor data to determine the performance characteristics of the UNEP pod.
Examination of the processed data files showed that NO2 and SO2 concentrations were being
reported as either negative or near zero concentrations when true environmental conditions were
not consistent with these values. The US EPA reported these findings to the UNEP staff early in
the monitoring process, but no conclusions were made about the cause of this apparent gas
measurement discrepancy. Once the monitoring was completed, the US EPA revisited the issue
and considered that the source of the problem might be in the executable macro provided by
UNEP, rather than in the sensors themselves. The algorithms and supporting parameters used in
the macro code to compute the NO2 concentrations, as well as the macro code itself, were
examined in detail. Some errors were discovered in the code, suggested revisions were made,
and the data were reprocessed using the updated code. The details of this process are documented
in the Appendix. The results of this detailed review were promising except for a signage
problem, the cause of which remains undetermined. It was decided that gas phase data from this
evaluation would not be incorporated into the results defined by this report. After the field tests
were concluded, UNEP staff were provided copies of all raw UNEP pod data as well as
collocated reference monitoring data to assist them in further elucidation of these issues.
Therefore, only meteorological (RH and temperature) and particulate matter data (PM2.5 and
PM10) from the UNEP pod were directly compared to the collocated reference monitors' data.
This report defines the specifics of the environmental test conditions used in the evaluation
(systems and conditions), data observations, summarization of key performance evaluation
findings, and ease of use features concerning the UNEP pod.
3

-------
2.0 Materials and Methods
2.1 Instrumentation
The UNEP pod was composed of an Optical Particle Counter for PM2.5 and PM10 (AlphaSense-
OPC-N2) measurements, two gas phase sensors for SO2 and NO2 (Alphasense model B-4)
measurement, a Global Positioning System (Sparkfun Venus GPS module and ANT 555 active
GPS antenna), a temperature and humidity sensor (Sensirion SHT21), and a Texas BeagleBone
Black. The latter is a single board computer that doubles as both the microprocessor and the
system control module. The US EPA did not open the encasement of the pod, and therefore other
components that might have been present (e.g., exhaust fans) but not immediately apparent are
not named here. The prototype received for evaluation had previously been operated in Nairobi,
Kenya to map the city's air pollution hotspots while also informally evaluating its reliability and
robustness.
The gas sensors incorporated into the UNEP pod typically come with some degree of
factory/manufacturer calibration. The full extent of such calibration/audit is unknown. The US
EPA's experience suggests that the gas phase sensors would have undergone laboratory
calibration by the manufacturer (zero and span check at temperatures ranging from -30 to 50 °C).
Information found on the manufacturer's website would appear to indicate that batch-to-batch
variability of the sensors is used to develop processing algorithms (Alphasense application note
AAN 803-02 - September 2016). Correction algorithms are often used to compensate for known
environmental artifacts (temperature, relative humidity, and interfering gasses) in field
measurements. Depending upon the age of the gas phase sensors, the value of the manufacturer's
original calibration is uncertain (Williams, 2017a). If the sensors were relatively young (e.g., <
1-2 months of age), then the manufacturer's calibration algorithm would be expected to be of
value. The age of the gas phase sensors in the UNEP pod was unknown at the time of the
evaluation, and the US EPA cannot further elaborate on this topic. The UNEP pod was used "as
is" without any attempt to conduct direct calibration of the gas phase sensors.
The OPC-N2 particulate matter sensor, in like manner, was used without any secondary
calibration performed by the US EPA. The OPC-N2 collects a total of 16 size-defined bin
designations of particle counts from 0.38 to 17 microns to which a proprietary manufacturer's
algorithm (Alphasense, 2013) is applied to develop mass concentrations (|ig/m3) for two size
fractions (PM2.5 and PM10). The processing algorithm shared by the UNEP provided for US EPA
to have access to the PM2.5 and PM10 mass concentration data from the device (including 16 size
bins). The US EPA did not attempt to investigate the size bins and how they were being used to
establish the various PM size fraction mass densities. It is believed that the AlphaSense OPC-
N2's algorithm for converting size bins into mass density was being used. The size bin
integration algorithm is typically proprietary and investigating it was beyond the scope of this
research effort.
The US EPA evaluated the temperature and relative humidity sensors included in the UNEP pod
with no additional calibration. Data from these two sensors were acquired as part of the raw data
output from the Beaglebone and processed as defined later into their final report state.
4

-------
The US EPA operates the AIRS test platform on its Research Triangle Park campus (Williams,
2014b). A GRIMM Technologies, Inc. (Douglasville, GA) Class IIII designated PM2.5 Federal
Equivalent Method (FEM) monitor is under continuous operation at that site. The GRIMM
monitor also provides PM10 mass concentration estimates but is not a US-designated FEM for
this aerosol mass size fraction. The European Union, however, has designated this monitor as an
equivalent reference monitor meeting the EN12341 standard
(http://ec.europa.eu/environment/air/qualitv/ legislation/pdf/equivalence.pdf). and it is used herein
as such to provide comparison data for informational purposes. The GRIMM monitor was used
because it was already fully operational at the AIRS, required no additional EPA resources for its
operation, and provided the ability to examine 1-minute data collection periods from the pod. No
other reference PM monitors were available for this effort. That alternative FRM/FEM monitors,
such as those involving regulatory filter-based monitoring or even the Tapered Element
Oscillating MicroBalance (TEOM) approach, might have provided additional benefit was
recognized. However, no such methods were available during this study.
A cavity attenuated phase shift (CAPS) NO2 analyzer was operated at the AIRS. Specifics of the
description and basic operation of the model T500U CAPS NO2 analyzer (Automated Equivalent
Method: EQNA-0514-212) are available elsewhere (EPA, 2014). Other instrumentation included
a Teledyne T500U NO2 analyzer and a Thermo 43C SO2 analyzer. While there were no FRM or
FEM temperature or relative humidity (RH) monitors, reference data for these measures were
provided at a height of 3 meters at the AIRS using an R.M. Young 41382VC environmental
probe, a widely accepted research grade device. These established methods are covered under a
QAPP (Alion, 2013). Specifically, all gas analyzers undergo automated daily zero, span, and
quality control checks. The GRIMM monitor's optics are calibrated annually by the
manufacturer, with its flow rate verified on a quarterly basis. The response of the meteorology
sensor undergoes annual audit. Reference data were available for the time frame of the sensor
evaluation as 5-minute averages. Study staff reviewed all raw reference data to ensure the
various systems were operating in a nominal fashion within the guidelines of the QAPP noted.
Only validated data were retained and used in the analysis. Photographs of the GRIMM monitor
and selected reference monitors used in this evaluation are shown in Figure 1.
5

-------
Environmental Dust Monitor
Model 190
GRIMM Aerosol Technik
Figure 1. GRIMM monitor (left) and select reference monitors at the AIRS
2.2 Deployment
The UNEP pod was deployed at the AIRS facility on November 3, 2016 for approximately I
month (33 days). Figures 2 through 4 depict the UNEP pod system as well as the reference
monitoring station. Figure 2 provides a close-up of the UNEP pod installed inside the
environmental shelter on the test platform at the AIRS, with the reference monitoring inlets in
the near background. The two monitoring stations (UNEP pod and reference monitors) were
within an inlet distance of 10 meters. The UNEP pod was housed in the aluminum shelter (1.11 m
0.91 i ,94 meters). This enclosure ensured sensor protection from both windblown rain and
direct sunlight while allowing unimpeded airflow. Specifics concerning the aluminum shelter
have been previously reported (Jiao, 2016). The UNEP pod was placed in the center of the
middle shelf at a height of 1.07 meters from the ground in an upside-down horizontal
configuration to provide unrestricted access to ambient air to its inlets/outlets, presumed to be
around the thin edges of the encasement. The UNEP pod was connected to 110V AC power,
which was available through the power strip housed on the lower shelf. Data were logged to the
internal microprocessor continuously and subsequently downloaded to a dedicated laptop
computer via a direct (wired) connection weekly. As shown in Figure 2, the UNEP pod was
oriented with its inlets (presumed to be through the thin edge of the encasement) fully open with
respect to the ambient air plenum. Telephone conversations as well as direct inspection by a
representative of the UNEP team who viewed the site early in the collocation process provided
full agreement that this orientation was appropriate. The shelter was kept locked except during
data recovery periods to ensure data integrity status. The US EPA is a secure facility, with
6

-------
restricted public access, and the only known external visitor to the site (UNEP representative)
was under direct staff escort.
Figure 2. Orientation of UNEP pod and associated wiring connections
Figure 3. The aluminum test shelter with the reference monitor in the background
7

-------
Figure 4. Select AIRS reference monitor inlets
At the time of deployment, the time on the BeagleBone was set to EST using PuTTY with the
following command: root@beaglebone: -# date —set "DD MM YYYY HH:MMam/pm"
2.3	Data Collection Procedure
Communication between the UNEP pod and the study computer used software mandated by the
UNEP. The UNEP pod operated with BeagleBone serial-over-USB drivers, PuTTY, a 64-bit
PuTTY implementation of SSIT and Telnet for Windows and Unix platforms, along with an
Xterm terminal emulator. A tool for transferring files securely between computers using an SSH
connection (psftp) was also utilized. The software application psftp was loaded onto the study
computer, which used Microsoft Office Windows 7. The computer was used to set the date time
stamp of the UNEP pod's internal clock at the start of data collection and weekly thereafter, to
store data downloaded from the UNEP pod's microprocessor, and ultimately to process the raw
data into a completed format. Details about setting up these programs are provided in the UNEP
AQ Monitoring Unit User Manual. Specifics about the use of these key software applications are
provided as part of the Appendix.
2.4	Data Processing
Data were processed following each weekly download using software programs (executables)
provided by the UNEP. Initially, these software applications were used as received without
change. As reported later in this document (see Appendix), corrections were made to one of the
scripts to overcome a problem we observed. The two executables consisted ofMerge_CSV.cmd
and Gas_ppb.xlsm.
8

-------
The Merge.CSV executable needed to be run once for each sampling day. The resulting file
combined 24 individual hourly files into a single 24-hour daily file called "combined.csv". The
hourly files as well as the combined.csv file contained raw sensor data. In the case of PM, these
files contained the binned distribution of PM, which is the generic output of the OPC-N2. In
brief, the data processing steps included the following:
•	Activating the Merge CSV.cmd application and then applying it to each hourly data
set to yield a file called "Combined.csv".
•	Saving the Combined.csv file as the .xls (e.g. 1 l-07-16.xls) to give it date
stamp recognition for record keeping.
The Excel macro Gas_ppb.xlsm was then executed on the .xls file. This macro converted
all sensor voltage data into concentration data. An example of excerpts from the raw and
processed files for the date 11-07-16 are provided in Figure 5 and is continued in Figure 5a (for
viewing purposes).
0
«3-
File
1
,:A
&
IF! -
ID
Paste
V
;	ll-07-16.xls [Compatibility Mode] - Excel
Insert Page Layout Formulas Data Review View Developer 0 Tell me what you want to do
| Calibri
B I U
"11 * A A — =
I m - <0 - ^ - =: =
= ' Sf'Wrap Text	General
^ | «= *= !| Merge & Center * $ * %
Alignment	I Numi
Srivastava, Manu E
1 IPS
.00 Conditional Format as Cell
-*.o
Formatting* Table* Styles*
ft	Styles
ffl et
Insert Delete Format
5] AutoSum -	O
S] Fill" Z
Sort & Find&
¦f Clear* Filter* Select*
Editing

A
B
c
D
E
F
G
H
'
J
K
L
M
N
o
P
Q R
s
l
Date
SysTime
Temp
RH
Dewpoin
PM1
PM2.5
PM10
SO2AIN0
S02AIN1
N02AIN2
N02AIN3 ECl_AE_rr
ECl_WE_r
EC2_AE_rr
EC2_WE_r Tc_S02 Tc_N02
WEc_S02 \
2
11/7/2016
0:00
13.58
60.37536
6.084073
130.3696
134.6819
134.8274
0.332095
0.341675
0.226274
0.222407
46.095
65.675
-49.726
-55.726
-2.568 0.8568
68.07279
3
11/7/2016
0:01
13.54
60.39528
6.042136
45.1022
47.63704
51.38897
0.330029
0.340576
0.22689
0.222363
44.029
64.576
-49.11
-55.11
-2.584 0.8584
66.98979
4
11/7/2016
0:02
13.52
60.33102
5.998447
44.78125
46.80687
47.52411
0.330293
0.339038
0.226582
0.222539
44.293
63.038
-49.418
-55.418
-2.592 0.8592
65.45979
5
11/7/2016
0:03
13.5
60.3904
6.012698
41.58971
43.59165
44.69676
0.330161
0.340444
0.227856
0.222012
44.161
64.444
-48.144
-54.144
-2.6 0.86
66.87379
6
11/7/2016
0:04
13.49
60.38714
5.993072
41.81435
44.24715
46.53338
0.330513
0.34207
0.22667
0.223066
44.513
66.07
-49.33
-55.33
-2.604 0.8604
68.50379
7
11/7/2016
0:05
13.5
60.33102
5.998447
43.67346
46.04177
48.80359
0.329853
0.341059
0.226802
0.222495
43.853
65.059
-49.198
-55.198
-2.6 0.86
67.48879
8
11/7/2016
0:06
13.49
60.29971
5.981504
42.94137
45.10302
47.47677
0.328799
0.339917
0.227461
0.222495
42.799
63.917
-48.539
-54.539
-2.604 0.8604
66.35079
9
11/7/2016
0:07
13.48
60.35746
5.966325
44.32006
46.86533
50.75589
0.331611
0.340884
0.226626
0.222407
45.611
43.194
64.884
64.049
-49.374
-49.154
-55.374
-55.154
-2.608 0.8608
-2.608 0.8608
67.32179
66.48679
10
11/7/2016
0:08
13.48
60.35908
5.978823
42.62786
45.08301
47.04244
0.329194
0.340049
0.226846
0.222934
11
11/7/2016
0:09
13.51
60.50911
6.021524
43.5279
45.67579
49.92786
0.329897
0.340532
0.226758
0.222583
43.897
64.532
-49.242
-55.242
-2.596 0.8596
66.95779
12
11/7/2016
0:10
13.45
60.61959
6.000828
42.18829
44.68388
46.7774
0.33104
0.339829
0.226802
0.223374
45.04
63.829
-49.198
-55.198
-2.62 0.862
66.27879
13
11/7/2016
0:11
13.44
60.5834
5.969734
42.74089
45.26595
47.1226
0.329678
0.340488
0.227241
0.222803
43.678
64.488
-48.759
-54.759
-2.624 0.8624
66.94179
14
11/7/2016
0:12
13.45
60.49768
5.955531
45.40805
48.0681
50.62657
0.331479
0.340532
0.22645
0.222934
45.479
64.532
-49.55
-55.55
-2.62 0.862
66.98179
15
11/7/2016
0:13
13.42
60.43998
5.948423
44.29506
46.5328
48.73898
0.330864
0.341147
0.226758
0.222275
44.864
65.147
-49.242
-55.242
-2.632 0.8632
67.60879
16
11/7/2016
0:14
13.4
60.49115
5.916274
42.68715
45.36761
52.13263
0.329634
0.339697
0.226186
0.221836
43.634
63.697
-49.814
-55.814
-2.64 0.864
66.16679
17
11/7/2016
0:15
13.4
60.28669
5.886076
41.09535
43.27734
46.55693
0.331523
0.340488
0.226055
0.222847
45.523
64.488
-49.945
-55.945
-2.64 0.864
66.95779
18
11/7/2016
0:16
13.36
60.4566
5.896645
41.5319
43.54481
44.94678
0.33082
0.339917
0.226538
0.222143
44.82
63.917
-49.462
-55.462
-2.656 0.8656
66.40279
19
11/7/2016
0:17
13.36
60.15981
5.818287
42.40712
44.65121
46.12545
0.328887
0.340576
0.226934
0.222275
42.887
64.576
-49.066
-55.066
-2.656 0.8656
67.06179
20
11/7/2016
0:18
13.34
60.27531
5.807595
42.2036
44.40606
45.38359
0.330732
0.340356
0.226494
0.222495
44.732
64.356
-49.506
-55.506
-2.664 0.8664
66.84979
Figure 5. Example of processed file
9

-------
ll-07-16.xls [Compatibility Mode] - Excel
Review View Developer 0 Tell me what you want to do
Srivastava, Manu H
\=p Wrap T ext
§ Merge & Center -
lment ft
General *
t . o/ , «-.0 .00
> " TO * .00 -».0
Number ft
m w ^
Conditional Format as Cell
Formatting" Table" Styles"
Styles
Insert Delete Format
Cells
T AutoSum "
El Fill -
Sort& F
Clear- Fi|ter. j,
Editing

S
T
U ,
V
W
X
Y |
Z |
AA
AB
AC
AD
WEc_S02
WEc_N02
S02_ppb
N02_ppb




Sensor
N02-B43F
S02-B4

68.07279
-13.1208
0.142114
-0.02517




Algorithm
1
4

66.98979
-12.954
0.139853
-0.02485




Temp
nT
k"T

65.45979
-12.9581
0.136659
-0.02486




-30
1.3
-4

66.87379
-12.7402
0.139611
-0.02444




-20
1.3
-4

68.50379
-12.8865
0.143014
-0.02472




-10
1.3
-4

67.48879
-12.8877
0.140895
-0.02473




0
1.3
-4

66.35079
-12.776
0.138519
-0.02451




10
1
-4

67.32179
-12.8729
0.140547
-0.0247




20
0.6
0

66.48679
-12.8422
0.138803
-0.02464




30
0.4
20

66.95779
-12.9136
0.139787
-0.02478




40
0.2
140

66.27879
-12.7893
0.138369
-0.02454




50
-1.5
450

66.94179
-12.7092
0.139753
-0.02438








66.98179
-12.8379
0.139837
-0.02463








67.60879
-12.7363
0.141146
-0.02443








66.16679
-12.7747
0.138135
-0.02451








66.95779
-12.7925
0.139787
-0.02454








66.40279
-12.6477
0.138628
-0.02426








67.06179
-12.5945
0.140004
-0.02416








66.84979
-12.614
0.139561
-0.0242








Figure 5a. Example of processed file (continued)
2.5 Data Processing Observations
Data from the UNEP pod were recovered and processed on a weekly basis. The US EPA
believed that the UNEP pod was operating nominally as the study effort continued. That is, data
were being logged, given time/date stamps, and stored in specific storage registers that were
specific to each hour of the collection period. The US EPA performed a cursory review of each
week's data to ensure that the pod was operational and that data were being logged. Raw data
was not extensively examined for completeness or quality assurance purposes on a weekly basis.
When all data collections were complete, the US EPA initiated a thorough review of all raw 1-
hour data records, as well as the Combined.csv files produced from the application of the
Merge_CSV.cmd file and the Excel files produced from the application of the Gas_ppb.xls
macro. During this review of the full data record, the US EPA discovered that some data records
corresponded to the wrong month. The problem occurred at the time when data collection moved
into a new month. It initially appeared that data records were being assigned a date stamp
corresponding to the previous month (November) rather than the current month (December).
This raised concern that data were being overwritten, but further inspection revealed that the
previous month's data record was being retrieved in addition to the current month's data when
data were collected for the same date of each month (Appendix Figure A8). It appears that the
data retrieval script and procedure is not specific enough to handle two different months. The
directory structure assumes all data are collected in a single calendar month, so no monthly
identification is specified. This problem affected all December data that had days of the month
that matched those in the November data set. The US EPA suspended sampling early in
December, but it is suspected that all the December data would have suffered this same re-
reporting of the earlier month's data.
10

-------
The same problem was observed in one of the November data downloads, wherein a few data
points from September were included with the November data with the same dates, even though
the US EPA did not record any data in September (Appendix Figure A9). These data looked very
different from the November and December data and are likely data recorded by UNEP prior to
this study that were still retained in the processor's memory.
During the collection period, the UNEP pod was not restarted and no data were deleted.
However, restarting the UNEP pod (powering on/off) as well as conducting a routine (scheduled)
data file deletion effort might not be the best approach to address the issue of re-reported data.
The best way to address the problem might be to revise the data download code and procedures
to eliminate the issue for future applications. The data directory structure in which the data are
stored needs to have separate subdirectories created for each month. The collection days (1-31)
would then be listed as subdirectories under each month. The directory structure currently has no
such monthly designations.
2.6 US EPA Quality Assurance Review and Application
As previously stated, US EPA conducted an intensive review of the raw and processed data
following the conclusion of data collection. Collocated reference monitoring data were reviewed
by US EPA for compliance with the QAPP requirements of the AIRS test platform and then used
without variation to yield a comparison database into which the UNEP pod data was integrated.
The UNEP pod recorded raw data every minute, whereas some of the reference monitors, (e.g.,
the GRIMM monitor), recorded data every 5 minutes. Therefore, the first order of processing
was to develop the means (steps) to allow for appropriate (matched) time intervals to be
examined for comparison. Data were excluded/manipulated in some fashion if they met any of
the following criteria:
•	The data from the 15-minute period before and after the weekly servicing visit from
the US EPA staff were excluded. Such disruptions have been shown to influence local
air quality associated with particle resuspension near the monitor intake (such as
would be the case here with the UNEP pod).
•	Data collected during the weekly automated calibration of the reference monitors
were excluded.
•	Data spikes in the UNEP pod data record, indicative of some systematic condition
that occurred consistently at the top of each clock hour, were excluded. The causal
agent for this UNEP pod artifact is unknown. An example of this artifact has been
provided in the Appendix.
•	An average was computed ONLY when valid data points were present at least 90% of
the time.
o To calculate 5-minute averages for the UNEP pod, 100% (five 1-minute data
records) of the raw data must be valid,
o To calculate 1-hour averages for the UNEP pod, > 54 1-minute data records
must be valid.
o To calculate 12-hour averages for the UNEP pod, > 648 1-minute data records
must be valid.
11

-------
o To calculate 24-hour averages for the UNEP pod, > 1296 1-minute data
records must be valid.
Additionally, while the UNEP pod never recorded an RH > 95%, an RH value known to often
impact light scattering PM sensors, the ambient reference monitor recorded multiple instances of
such events. To examine the impact of RH on UNEP pod PM response, UNEP pod PM data was
parsed based upon the reference RH value > 95% and this subset of the data was investigated
separately.
An automated executable macro was developed by the US EPA to ensure reliability in the
matching of date/time stamp records from reference and UNEP pod data files. The automated
process streamlined the full evaluation of the data to allow summary comparison at a wide
variety of time integration values (e.g., 5-min, 1-hour). Visual highlights within the spreadsheet
occurred when a data cell was empty, nonsensical (e.g., an alphanumeric instead of a value), or
otherwise outside the quality assurance requirements. Ultimately, processed and validated data
from both the AIRS reference and UNEP pod were integrated into a single electronic spreadsheet
having a matched date/time stamp, as shown in Figure 6.

A
B
C
D j
E 1
F |
G |
i	H 1
I ,
J |

1
UN Pod Avg. Date Time
UN Pod Temp C°
UN Pod RH % UN Pod PM2.5 ug/m3
UN Pod PM10 ug/m3 REF Avg. Date Time Grimm PM2.5 ug/m3 Grimm PM10ug/m3 3m Temp C
3m RH %

2
11/5/2016 0:00
22.52927157
49.07473626
70.35744451
78.6262086
11/5/160:00
9.007638889
11.58159722
15.68854167
71.42014

3
11/6/2016 0:00
17.67168198
47.90261947
63.74783816
66.47208909
11/6/160:00
6.455902778
8.147916667
10.03541667
74.85417

4
11/7/20160:00
18.49595053
47.360514
20.94406669
22.43103369
11/7/16 0:00
7.523263889
9.16875
10.75381944
74.65972

5
11/8/2016 0:00
16.98269258
46.02917471
18.42235046
20.12195629
11/8/160:00
5.184375
7.092708333
9.477083333
71.52431

6
11/9/2016 0:00
17.07373759
47.96910923
14.07029948
16.12286472
11/9/160:00
6.299652778
8.829861111
9.717013889
74.77431

7
11/10/2016 0:00
18.3130318
48.88996596
33.60964343
36.78001317
11/10/160:00
13.42951389
16.48229167
12.07291667
70.66667

8
11/11/2016 0:00
16.96827187
42.39615831
13.72703035
14.88755125
11/11/160:00
3.855208333
5.160416667
9.634027778
64.19097

9
11/12/20160:00
18.23010601
41.50084177
13.176741
16.07735437
11/12/160:00
8.873611111
13.32222222
11.159375
62.30208

10
11/13/2016 0:00
14.58621641
39.27744409
17.24384407
18.18480618
11/13/16 0:00
3.75
5.097569444
7.239583333
59.375

11
11/14/20160:00
14.00221908
41.16772463
25.02966162
26.02558612
11/14/160:00
8.156944444
9.404861111
6.310416667
64.65278

12
11/15/20160:00
15.81930035
62.70180348
44.81773015
48.00633315
11/15/160:00
12.80243056
14.72743056
9.169791667
95.05208

13
11/16/2016 0:00
16.8734417
50.86475816
27.37748779
34.61061959
11/16/16 0:00
21.81041667
25.925
8.994444444
81.72569

14
11/17/2016 0:00
17.11023355
49.71860786
21.13192672
23.92399093
11/17/160:00
30.10798611
34.790625
9.422569444
78.89931

15
11/18/2016 0:00
17.99556277
48.67458025
20.94949515
24.22557573
11/18/16 0:00
20.84703833
26.05609756
10.3261324
76.72125

16
11/19/20160:00
19.80849965
49.76366254
21.53732865
25.17405311
11/19/160:00
49.05625
55.9125
11.97986111
78.91319

17
11/20/2016 0:00
17.73521555
45.08251185
17.69548552
21.27294927
11/20/160:00
29.37708333
34.09618056
11.95763889
64.22569

18
11/21/2016 0:00
11.67861484
30.12890683
8.114227407
8.886371557
11/21/16 0:00
5.976388889
7.488194444
5.288194444
42.22569

19
11/22/2016 0:00
10.57931449
34.79296253
4.163992966
4.782626181
11/22/160:00
5.628819444
7.177777778
3.622916667
52.57292

20
11/23/2016 0:00
11.19658657
34.96437915
14.32285934
15.07641929
11/23/160:00
5.816319444
7.876041667
3.490972222
54.80208

21
11/24/20160:00
12.64923355
39.95206245
9.766703933
11.63295591
11/24/160:00
13.52256944
19.75625
5.536805556
61.10417

22
11/25/20160:00
19.39796886
52.74904584
13.14829628
15.01945683
11/25/160:00
19.16076389
21.49131944
12.22083333
81.71528

23
11/26/2016 0:00
22.65170092
53.02681887
18.4163077
21.53278171
11/26/160:00
19.87777778
22.190625
15.23888889
81.81597

24
11/27/2016 0:00
16.99722261
39.82199648
23.46660613
24.51347688
11/27/160:00
10.32256944
11.09305556
10.44861111
56.35417

25
11/28/2016 0:00
11.85524752
43.75764888
16.38422544
16.71468377
11/28/16 0:00
8.335416667
8.765972222
3.873611111
70.94792

26
11/29/20160:00
13.81388693
50.22238105
17.16757034
19.34154303
11/29/160:00
12.77676056
15.84683099
7.379929577
75.40493

27
11/30/2016 0:00
23.82958304
59.73180544
15.53505001
27.85076851
11/30/16 0:00
8.643356643
10.38811189
18.85
80.18881

28










Figure 6. Matched time stamp data for UNEP Pod and AIRS reference data
12

-------
3.0 UNEP Pod Results and Discussion
The comparisons of PM2.5, PM10, RH, and temperature data measured by the UNEP pod and
collocated AIRS reference monitors are reported here. As previously stated, while gas phase
pollutants were measured during the evaluation, a comparison to reference monitoring data was
not pursued. This comparison was not pursued because of a lack of reasonableness in the data
with respect to observed reference monitoring concentrations over the course of the study. The
UNEP has been provided raw data from its pod as well as reference monitoring data to allow
them to further investigate gas phase sensor response. Some observations are noted in the
Appendix, but no further discussion concerning gas phase sensor performance is available in this
current report.
In this study, 1-minute data from the UNEP pod were averaged over various integration periods
(e.g., 5-minute, 1-hour, 12-hour, and ultimately 24-hours) and then compared with the time/date
stamp matched reference data. These comparisons included time series inspection as well as
linear regression statistical analyses. Data associated with the monitoring period from 11/03/16
through 12/02/16 were available for these comparisons, with a total of 40264 1-minute records
collected. Ultimately, a total of 39508 1-minute records was included following full execution of
all QA requirements. This equated to a 98.0% 1-minute data inclusion rate available for
statistical treatment. Examination of the impact of RH on pod performance resulted in a higher
exclusion rate of the data, resulting in only 91.3% of the original data records being available for
analysis.
Various parameters were theorized as having a potential impact on the response of the UNEP
pod with respect to reporting PM mass concentrations. More specifically, the impact of both
ambient RH and temperature were investigated and reported for both PM mass fractions.
3.1 RH Comparison
A linear relationship between the UNEP pod's measured RH data and the AIRS reference RH
data was observed (R2 = 0.91). The slope (m=0.61) in Figure 7 indicates that while there was a
linear response, the UNEP pod significantly underreported the true ambient RH (by ~ 40%). The
time series plot shown in Figure 8 clearly depicts this response issue. It should also be noted that
the UNEP pod had a positive bias of- 3% (intercept of the linear equation). In the US EPA's
previous experience, it is unusual for most commercially available RH sensors to deviate this far
from the true value if they are performing correctly. It is theorized, in this case, that the RH
sensor resides inside the encasement of the pod, and therefore the readings it provides might be
impacted by non-ambient conditions. In other words, there might be a drying effect of the
interior air space inside the encasement of the pod, effectively lowering the reported RH. Heat
emanating from the internal sensors might be such a drying mechanism. An alternative
explanation is that the RH sensor might be positioned in a less than favorable fashion within the
pod (i.e., next to something warm) or its original calibration response algorithm was less than
adequate. Removal of the sensor from the UNEP pod for testing under standalone conditions
would be one means of examining this question.
13

-------
Daily Comparison of the UN Pod with Reference
RH
70
60
S? 50
S 40
¦a
S. 30
z
=) 20
10
0
0	20	40	60	80	100
Reference RH, %
Figure 7. Linear Regression of the UNEP pod versus the AIRS reference RH
100
80
so 60
^ 40
20
0
11/1/

Daily RH vs Time Comparison


•
m ~ ^
mm m.


••••••
•• ••
• • •
•
m
• •
• •


* •A - a • • -




— w ~











'2016 11/6/2016 11/11/2016 11/16/2016 11/21/2016 11/26/2016 12/1/2016 12/6/2016
• Reference RH • UN Pod RH
Figure 8. Time series showing the response offset of the UNEP pod versus the AIRS reference RH
It is important to have RH data that are accurate and coincide with other environmental
measures. The US EPA has found optical-based PM low-cost sensors frequently to be highly
sensitive to RH values exceeding 95% (Jiao, 2016). Sensors of these types often yield extremely
biased (high) responses. Exclusion of sensor data above the 95% response inflection point has
been shown to significantly improve performance comparisons versus reference monitors
(Williams, 2014b). Therefore, the US EPA investigated the impact of RH relative to the PM2.5
and PM10 responses to determine whether an additional quality assurance parameter needed to be
included as part of the raw data exclusion criteria. True (ambient) RH data obtained from the
collocated reference monitor were used in these investigations because of the clear


y = 0.6119x+3.2029 <
R2 = 0.9143 _



P
f	


_



A	

















14

-------
underreporting of the UNEP pod's RH sensor. Time series comparison of ambient RH with
respect to PM2.5 and PM10 responses is shown in Figure 9. As can be seen, the RH and PM mass
concentrations follow no clearly observable pattern.
90
80
70
60
mE 50
40
30
20
10
Ct
Daily PM Content and RH Time Series
100
















90








80


















70







Cd
60

I / \

/





A
7\ j
/



50







ACi
11/1/2016 11/6/2016 11/11/2016 11/16/2016 11/21/2016 11/26/2016 12/1/2016 12/6/2016
	GRIMM PM2.5 	UN POD PM2.5 	 Grimm PM10 	UN Pod PM10	Reference RH
Figure 9. PM response relative to ambient RH conditions
Regression analysis of this same 24-hour average data set reveals a modest trend of an increasing
UNEP Pod PM response with an increasing ambient RH, as shown in Figure 10. Even so, the
regression outcome was not highly correlated sufficiently (R2 < 0.18) to warrant data exclusion
for this potential co-factor. Examination of shorter averaging time intervals (e.g., 5-minute, 1-
hour) as reported in Table 1, and illustrated in Figures 11 and 12 using 1-hour average data,
revealed even less statistical basis for RH exclusion.
15

-------
UN Pod PM Content Regressed with Reference RH
90
80
mE 70
~5o 60
£ 40
o
a- 30
Z
=> 20
10
0
40	50	60	70	80	90	100
Reference RH, %
• UN Pod PM10 • UN Pod PM2.5 	Linear ( UN Pod PM10) 	Linear (UN Pod PM2.5)
Figure 10. Regression of 24-hr average PM concentration from the UNEP pod versus reference RH
Table 1. Impact of applying > 95% RH exclusion criteria on the strength of the UNEP pod versus
reference PM mass concentration comparison (reported as R2 values) at various averaging times
Time Interval
R2 PM2.5 vs Reference RH
R2 PM10 vs Reference RH
5 minutes
0.02
0.02
1 Hour
0.02
0.02
4 Hours
0.02
0.02
12 Hours
0.02
0.03
24 Hours
0.14
0.18









•
* t





















y = U.bl4oX - 1 /. /44



•


K
= U.181

t
		
		





•
J.	
		
ill
			
	 m»M"
r>»1
	



.1
y = 0.5103X-13.27

•	
•

w	
R2:
= 0.1442

16

-------
UN Pod 1 Hour Averaged PM2.5 Regressed
with Reference RH
100
ao
3
80
40
20





rr$
	5__	 m
y = 0.1192x + 13.6 #
d2 _ n ni -7/i
••
m
fl
•


l&g
«*j
S
i oirjrt
*
•
Jk
(j2
to:
¦•J*


10 20 30 40 50 60 70 80 90 100
Reference RH, %
Figure 11. Regression of 1 -hr average PM2.5 concentration from the UNEP pod versus reference RH
UN Pod 1 Hour Averaged PM10 Regressed with
Reference RH
120
0
E 100
bio
3 80
D
rH
> 60
o 40
| 20
0
y = 0.1418x +15.146
R2 = 0.0219
10
40 50 60 70
Reference RH, %
100
Figure 12. Regression of 1-hr average PM10 concentration from the UNEP pod versus reference RH
3.2 Temperature Comparison
Figure 13 shows a time series plot of temperatures measured by the UNEP pod and the ambient
reference monitor. Regression of the comparison revealed excellent agreement between the
UNEP pod's temperature sensor's response versus reference data, with an R2 value > 0.96, as
shown in Figure 14. Nevertheless, the sensor revealed a significant degree of positive bias (> 7
°C). We believe that it is likely that this bias is related to sensor being embedded within the
encasement of the UNEP pod, therefore potentially being heated by its internal electronics. An
alternative causality might be, as noted previously the manufacturer's calibration or the raw
signal conversion algorithm developed for the sensor. As in the case of RH, the linear regression
of PM2.5 and PM10 versus the reference temperature, even on a 24-hour average basis, revealed
no statistical association, as shown in Figure 15.
17

-------
30
Daily Temperature vs Time Comparison
25
20
15
10
5
0
11/1/2016 11/6/2016 11/11/2016 11/16/2016 11/21/2016 11/26/2016 12/1/2016 12/6/2016
• Reference Temp • UN Pod Temp
Figure 13. Time series comparison of UNEP pod versus AIRS reference temperature
Temperature Comparability of the UN Pod with
Reference Monitoring


y = 0.9187x + 7.8952


R2 = 0.9648
^ 	
M		

^	V*


*••••











0	5	10	15	20
Reference Temp, °C
Figure 14. Regression of UNEP pod versus AIRS reference temperature
18

-------
UN Pod PM Response Regresssed with Reference Temperature
90
80
CO
70
60
5 50
£ 40
S. 30
H 20
10
0








y = 1.
9893x+ 6.1996

y= 1.4588x + 8.
4535


*2 = 0.1996
•
R
= 0.124]

••
6	8	10	12
Reference Temperature, °C
14
16
18
20
UN Pod PM10
UN Pod PM2.5
•Linear (UN Pod PM10)
¦Linear (UN Pod PM2.5)
Figure 15. Regression of 24-hr average UNEP pod PM response versus reference temperature
3.3 PM Mass Concentration Comparisons
As stated in previous sections of this report, neither temperature nor relative humidity comparisons
revealed any significant correlation association with the UNEP pod's PM response. Therefore, all
data meeting the inclusion criteria discussed in Section 2.5 were incorporated into the comparisons
reported herein, resulting in a very large data set of 1-minute measurements (39,508) available for
statistical review. As reported in many US EPA examinations of sensor performance, longer
averaging times (e.g., 24-hour average) typically result in improved statistical agreement between
sensors and reference monitors (Jiao, 2016). This result is directly relatable to the general
smoothing of data over the longer averaging times. However, the current report shares selected
findings associated with shorter time intervals because many elements of the public sector attempt
to use shorter periods of environmental monitoring (hours or even minutes) in conducting exposure
assessments. Since PMio and PM2.5 are often highly related with respect to mass concentration
trends, both mass concentrations are reported here in the general discussion of the UNEP pod's
performance characteristics. In general, findings from both mass fractions resulted in the same
pattern of UNEP pod response and performance characteristics. Figures 16 and 17 reveal 24-hour
average time series trends associated with both mass fractions. There are periods in both mass
fractions in which the UNEP pod either significantly over-reported or under-reported the true
(ambient) mass concentration. In numerous instances, the mass concentration difference between
the UNEP pod and reference monitor was more than 100%.
19

-------
Daily PM2.5 Time Series
80
70
m 60
£
"&> 50
3
40
LD
™ 30
20
10
0
11/1/2016 11/6/2016 11/11/2016 11/16/2016 11/21/2016 11/26/2016 12/1/2016 12/6/2016
• GRIMM PM2.5 • UN POD PM2.5
Figure 16.24-hr average PM2.5 concentration time series for the UNEP pod and AIRS reference monitor







1 •








•





•




•

> • •
> • • •




m _
• !



• • • • * •
•



• • •
w •

Daily PM10 Time Series







•






•








•





•





•
• • •




• -
• * • • • .


•
• • • t •


• s
# A

•'•' • •'

W • 9

11/1/2016 11/6/2016 11/11/2016 11/16/2016 11/21/2016 11/26/2016 12/1/2016 12/6/2016
• Grimm PM10 • UN Pod PM10
Figure 17.24-hr average PM10 concentration time series for the UNEP pod and AIRS reference monitor
The 24-hour average regression comparisons for both size fractions, depicted in Figures 18 and 19,
reveal no statistical association between the pod's response and collocated reference monitoring (R2
< 0.0002). This lack of agreement would not be expected to be related to a sensitivity issue of the
OPC-N2 based upon the manufacturer's specification data and the US EPA's own experience
operating this same PM sensor (EPA, 2017a). Ambient concentrations often exceeded 15 |ag/m3, a
value which should be easily detected with most optical particle sensors. Additionally, the AIRS
experienced a multi-day episode characterized by transported windblown forest fire smoke during
the evaluation that resulted in PM2.5 concentrations exceeding 50 |jg/m3. It should be noted that the
two PM2.5 and PM10 data points > 60 |igm3 observed on the first 2 days of the study warranted
additional review, as they appeared to be potential outliers.
20

-------
Daily PM2.5 Comparability of UN Pod with FEM
•
•












y:
= -0.0202X+ 22.744


•

R2 = 0.0002


•




	/•<
«•%
•	
•


• *
. •
>• *


•
•




0	10	20	30	40	50	60
GRIMM FEM, ug/m3
Figure 18. UNEP pod versus reference 24-hr average PM25 concentration
Daily PM10 Comparability of UN Pod with
GRIMM







•

y = 0.0183x + 25.1
D25
•


R
z = 0.0002








•





•
•



A.




:r» "
	
	#	


1 >
•w ¦




I i





0	10	20	30	40	50	60
GRIMM PM10, ug/m3
Figure 19. UNEP pod versus reference 24-hr average PM10 concentration
Figure 20 shows the 5-minute average data for the first half of the study only, to examine details
of the pod's behavior in the first 2 days of data reporting compared to subsequent days, as well
as to examine the relationship of the pod's PM to the reference RH. The pod's PM does appear
to make a sudden shift early on 11/6/16, likely explaining the high 24-hour average
concentrations in the first 2 days of the study. However, removal of these first 2 days of
monitoring before the downward baseline shift had minimal impact on the daily pod response for
both size fractions versus FEM comparisons (R2 < 0.07).
Considering their relationship to RH, the PM concentrations of the pod appear generally to track
ambient RH. The reason for the lack of correlation reported previously is the periodic drop in
21

-------
concentrations to zero or near zero, typically during or towards the end of high RH periods.
Further review of these same time periods in which data were excluded when the pod's PM2.5
response had fallen sharply to values between 0 and 5 jig/m3 provided minimal improvement of
the pod's correlation with the ambient FEM (R2 < 0.21).
5-Minute Averaged PM2.5 and RH vs Time
100
90
80
70
.£ 60
Ctf)
550
IN
5 40
Q_
30
20
10
m
iU If
1/1,
•	1 '• I
i : i !
•	• # I
•	| 1 •
fi A.,
t> r>.
f; i : I
i ! ! b%i
ft
I !
: f
• • •
I' |/
1 j
IV
•
•
: :
V w
s *
• • • •
m
• rW

* -i

L-v

• • i*•
aa
• L •
M

A*	1
A/vy
11/4/2016	11/6/2016	11/8/2016	11/10/2016	11/12/2016
• GRIMM PM2.5 • UN Pod PM2.5 • REF RH
11/14/2016
100
90
80
70
60
50
40
30
20
10
11/16/2016
Figure 20.5-minute UN EP pod response versus ambient RH at study onset
Regressions of 1-hour averages of UNEP pod concentrations versus reference PM concentrations
are shown in Figures 21 and 22. As these two figures show, and as Table 2 reports, no
observable association between the UNEP pod PM and the reference monitor PM existed for
either mass fraction.
22

-------
PM2.5 1 Hour Averaged Comparability of UN Pod
with FEM

100

90

80
m
70
,E
CuO
60
3

-O
50
O
Q.
40
z
30
3

20

10

0


iA _ u = n 14?y + ?n is

Jt ij
0074






•
•


m mm m — ¦	


•
•



20
40	60
GRIMM FEM, ug/m3
80
100
Figure 21.1-hour average UNEP pod versus AIRS reference PM2.5 concentration
PM10 1 Hour Averaged Comparability of UN Pod
with GRIMM

120

100
ro

,E
80
CuO

3

T3
60
O

Q.

z
40
3


20

0

•




J *

y = 0.1842x +22.251
R2 = 0.0135

&





c


^ *• •


>•» • _
	•	


H
p* r
• • • •


20	40	60	80	100
GRIMM PM10, ug/m3
120
Figure 22.1-hour average UNEP pod versus AIRS reference PM10 concentration
Table 2. Impact of averaging time on regression results comparing UNEP pod versus reference
PM mass concentration
Time Interval
UNEP PM2.5 vs Reference PM2.5
UNEP PM10 vs Reference PM10
5 minutes
y = 0.15x +20.05; R2 = 0.008
y = 0.19x +22.10; R2 = 0.01
1 Hour
y = 0.14x +20.16; R2 = 0.007
y = 0.18x +22.25; R2 = 0.01
4 Hours
y = 0.13x +20.49; R2 = 0.007
y = 0.17x +22.59; R2 = 0.01
12 Hours
y = 0.09x +21.09; R2 = 0.004
y = 0.12x +23.49; R2 = 0.008
24 Hours
y = -0.02x + 22.74; R2 = 0.0002
y = 0.02x + 25.03; R2 = 0.0002
23

-------
4.0 Ease of Use Features and Concerns
4.1	Hardware
The US EPA found the UNEP pod to be of solid construction and robust relative to its ease of
use once it was initialized (software acquisition and familiarity). The UNEP pod, in its normal
state, is to be mounted on a vertical structure. The US EPA operated the UNEP pod with the
primary encasement in a horizontal position, with the inlets to the sensors (presumed to be
around the narrow edge) in a vertical orientation (see Figure 2). UNEP consultation concluded
that this orientation would not obstruct the inlets.
It would have been beneficial for the US EPA, as an end user, to have had a greater
understanding of the UNEP pod and its components available through the labeling of the UNEP
pod encasement's parts. For example, defining the various inlets, fan housings, and other
features either on the UNEP pod itself (preferred) or in the provided study materials (operating
procedures) would have engendered greater confidence that all users could operate the device
successfully. One could envision other operators orienting the device in such a manner that an
inlet or other feature was hindered or impacted by its placement. Although the UNEP pod did
have indicator lights reporting its base state of operation, an actual LED screen on the UNEP
pod's face showing real-time data values would be beneficial for assuring basic operating status.
4.2	Data Processing
The UNEP pod was operated reliably, and data accessed successfully, with the exceptions
previously noted about processing errors. Processing the data from its raw form into its final
form was labor intensive. It would be valuable to other users if the developers could make the
following design improvements:
•	Incorporate automated processing using a script directly integrated into the
microprocessor that takes the raw data and performs all the data transformations
without operator involvement, allowing for the output file to be immediately useable.
The US EPA has developed numerous pod systems for collecting both particulate
matter and gas phase pollutants and such design features have proven to be extremely
effective.
•	Integrate a removable SD card. A removable SD card would allow processed data to
be easily accessed from the UNEP pod in situations where cellular communication is
not reliable or available.
•	Consider solar power with back-up battery solutions. The current design requires
access to land-based power supplies (e.g., 115 volt). It is doubtful that every remote
location will have such a benefit. In the US EPA's experience, relatively small (18
inch by 18 inch) solar panels are sufficient to power a pod of this nature.
24

-------
4.3 Study Limitations
Several important study design limitations need to be addressed relative to the observations
reported in this effort. A primary concern is the fact that only a single UNEP pod was evaluated.
The US EPA has observed in examining low-cost sensor performance that even when sensors are
manufactured as part of the same bulk process, a wide range in performance can occur under
various test scenarios (Jiao, 2016). Therefore, the poor association observed here with respect to
the OPC-N2 response, as well as the unknown factor(s) impacting the gas phase sensors, might not
be reflective of the UNEP pod's true performance if a larger number of pods had been evaluated.
Study resources limited the US EPA's ability to operate the UNEP pod for a longer period of
time during the evaluation. The UNEP pod's response was examined under conditions nominal
for the eastern U.S. geographical area during only a single seasonal period (fall). The site
experienced a fairly narrow range of 24-hour average temperatures (typically between 3 to
20°C), with some periods of extended precipitation (rainfall). Collectively, this suggests that
evaluation of the UNEP pod under different climatic conditions might yield different results.
Nevertheless, no statistical basis for either the RH or temperature to be a factor influencing the
pod's PM performance was observed.
The Research Triangle Park area typically has ambient PM2.5 mass concentrations well under 12
|ig/m3 (Williams, 2003). This location was selected not because it provided an expected wide
range in day-to-day variability in PM mass concentrations but because of its convenience (the
availability of US EPA staff and operating reference monitors). Tests conducted in a more
challenging environment might have yielded improved performance. The US EPA has witnessed
such improvements when PM sensors of the same type have been examined by other scientists
conducting evaluations where historically higher ambient levels of PM mass concentrations have
been reported (SCAQMD, 2017). Some low-cost PM sensors, such as the one examined here,
might have improved performance at the higher end of their operating range as opposed to values
at or near their lower detection limit. However, ambient concentrations observed during the
study did reflect a major emissions plume that significantly impacted local air quality for some
period of time (24-hour averages of PM2.5 > 50 ug/m3). Evaluation under even more extreme
conditions, such as those that might exist in many developing countries, may have resulted in
different performance characteristics.
The US EPA operated the pod using data recovery software developed by the UNEP specifically
for this study due to telecommunication restrictions (data security requirements). As such, data
were harvested weekly from the UNEP pod and then processed using executables provided. It is
unknown if the UNEP pod, when used in its normal state (data transmission via a
telecommunication service to a dedicated server with presumed automated data processing),
might have had an impact upon data quality.
The internal components of the UNEP pod were not examined, as the US EPA purposefully
tested the pod without any opportunity to negatively influence (damage) such components. It is
unknown if all internal components worked as the UNEP desired. A failure of various
components responsible for movement of air mass over or through the various sensor bodies
could have occurred without our knowledge.
25

-------
4.4 Conclusions
The US EPA's previous experience with low-cost sensor performance evaluations provided a
context in which to draw conclusions concerning the UNEP pod and its overall capabilities to
accurately estimate local environmental concentrations. The device was determined to be
extremely stable relative to its operational status. It worked without failure with respect to
collecting data for more than a 1-month period once it was initialized. The device appeared to be
solidly constructed (external encasement) and, to US EPA knowledge, no failure of primary
operating components (e.g., fans, electronic boards) occurred during the evaluation. The US EPA
speculates that the RH and temperature sensors within the UNEP pod are being influenced by
other electronic components, as observations indicate measurement values indicative of some
systematic influencing factor. This might present a significant issue if RH exclusion criteria were
ever applied to the data as a normal practice during UNEP pod operation.
26

-------
5.0 References
Alion Science and Technology. 2013. Quality Assurance Project Plan: PM and VOC Sensor Evaluation,
QAPP-RM-13-01(1), November 14, 2013. Research Triangle Park, NC.
Alphasense Air. Particulates. c2013. http://www.alphasense.com/index.php/products/optical-particle-
counter/. last accessed June 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. 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.
Kaufman A., Brown A., Barzyk T., and Williams. R. 2014. The Citizen Science Toolbox: A One-Stop
Resource for Air Sensor Technology. EM. 48-49.
South Coast Air Quality Management District (SCAQMD). 2017. Air Quality Sensor Performance
Evaluation Center, http://www.aqmd.gov/aq-spec. Accessed June 18, 2017.
U.S. Environmental Protection Agency (EPA). 2014. Ambient Air Monitoring Reference and Equivalent
Methods: Designation of Four New Equivalent Methods. Federal Register. 79: 34734-34735.
U.S. Environmental Protection Agency (EPA). 2017a. Air Sensor Toolbox for Citizen Scientists,
Researchers and Developers, https://www.epa.gov/air-sensor-toolbox. last updated April 6, 2017; last
accessed June 18, 2017.
U.S. Environmental Protection Agency (EPA). 2017b. Air Sensor Toolbox: Resources and Funding.
Resources from the Air Quality Sensor Performance Evaluation Center, https://www.epa.gov/air-
sensor-toolbox/air-sensor-toolbox-resources-and-funding#AOSPECenter. last updated April 6, 2017;
last accessed June 18, 2017.
U.S. Environmental Protection Agency (EPA). 2017c. Air Sensor Toolbox: Resources and Funding.
Resources from the European Commission Joint Research Centre, https://www.epa.gov/air-sensor-
toolbox/air-sensor-toolbox-resources-and-funding#EUCommission. last updated April 6, 2017; last
accessed June 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?dirEntryId=308114,
last accessed June 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?dirEntrvId=310285. last accessed June 18,
2017.
27

-------
Williams, R. 2014a. Findings from the 2013 EPA Air Sensors Workshop. EM. January: 5.
https://www.epa.gov/sites/production/files/2015-
06/documents/emmagazine.2013 air sensors wkshop.pdf. Accessed June 18, 2017.
Williams, R., A. Kaufman, T. Hanley, J. Rice, and S. Garvey. 2014b. 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://cfbub.epa.gov/si/si public record report.cfm?dirEntryId=297517.
last accessed June 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. 2014c. Sensor Evaluation
Report. U.S. Environmental Protection Agency, Washington,DC, EPA/600/R-14/143 (NTIS PB2015-
100611), https://cfpub.epa.gov/si/si public record report.cfm?dirEntryId=277270. last accessed June
18, 2017.
Williams R., Watkins T., Long, R. 2014d. Low-Cost Sensor Calibration Options. EM. January: 10-15.
Williams, R., Suggs, J., Rea., A., Leovic, K., Vette, A., Croghan, C., Sheldon, L., Rodes, Thornburg, J.,
Ejire, A., Herbst, M., Williams Sanders Jr. 2003. The Research Triangle Park particulate matter panel
study: PMmass concentration relationships. Atmospheric linvironment, 37: 5349-5363.
28

-------
6.0 Appendix
Data Harvesting and Processing Procedures
Before each data extraction, the current time on the BeagleBone was set by the study computer
using PuTTY using the following command:
roofchbeag/ebonedate
The results are provided in Figures A1-A5.
COM4 -PuTTY	1=1 @ J3_l
Debian GNU/Linux 3 beaglebone ttyGSO
3eagleBQard.org Debian Image 2016-01-24
Support/FAQ: http://elinux.org/Beagleboard:BeagleBoneBlack_Debian
UAQHI support patrmccormack@gmail.com
The IP Address for asbO is: 192.168.7.2
beaglebone login: root
Last login: Thu Sep 8 03:40:11 UTC 2016 on ttyGSQ
Linux beaglebone 4.1.15-ti-rt-r43 #1 SMP PREEMPT RT Thu Jan 21 20:13:58 UTC 2016
armv71
The programs included with the Debian GNU/Linux system are free software;
the exact distribution terms for each program are described in the
individual files in /usr/share/doc/"/copyright.
Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
permitted by applicable law.
root@beaglebone:~# date
Thu Nov 10 16:18:10 UTC 2016
root@beaglebone: ~# \_
Figure A1. PuTTY command code.
29

-------
$ COM4 - PuTTY 1 a 121
a :
Debian GNU/Linux 8 beaglebone ttyGSO

A
BeagleBoard.org Debian Image 2016-01-24


Support/FAQ: http://elinux.org/Beagleboard:BeagleBoneElack Debian


UAQHI support patrmccormac!c0gmail. com


The IP Address for usbO is: 192.168.7.2


beaglebone login: root


Last login: Thu Nov 10 16:18:05 UTC 2016 on ttyGSO


Linux beaglebone 4.1.15-ti-rt-r43 #1 SMP PREEMPT RT Thu Jan 21 20:13:53 UTC
2016

armv71


The programs included with the Debian GNU/Linux system are free software;


the exact distribution terms for each program are described in the


individual files in /usr/share/doc/*/copyright.


Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent


permitted by applicable law.


root0beaglebone:~# date


Thu Nov 17 10:05:54 UTC 2016


root@beaglebone: ~# [[

~
Figure A2. PuTTY command code (continued)
COM4 - PuTTY	I a 1 B
Debian GNU/Linux 8 beaglebone ttyGSO
SeagleBoard.org Debian Image 2016-01-24
Support/FAQ: http: //elinux. org/Beagleboard:3eagleBoneBlacfc_Debian
UAQHI support patrmccormacfc@gniail.com
The IP Address for usbO is: 192.163.7.2
beaglebone login: root
Last login: Thu Nov 17 10:05:50 UTC 2016 on ttyGSO
Linux beaglebone 4.1.15-ti-rt-r43 #1 SMP PREEMPT RT Thu Jan 21 20:13:58 OTC 2016
armv71
The programs included with the Debian GNU/Linux system are free software;
the exact distribution terms for each program are described in the
individual files in /usr/share/doc/*/copyright.
IDebian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
[permitted by applicable law.
root@beaglebone:~# date
IWed Nov 23 14:42:34 UTC 2016
root@beaglebone:|
Figure A3. PuTTY command code (continued)
30

-------
$ COM4 - PuTTY
Debian GNU/Linux 8 beaglebone ttyGSO
3eagleBoard.org Debian Image 2016-01-24
Support/FAQ: http://elinux.org/Beagleboard:3eagle3one31ack:_Debian
UAQHI support patrmccormack:@gmail. com
The IP Address for usbO is: 192.168.7.2
beaglebone login: root
Last login: Wed Nov 23 14:42:30 UTC 2016 on ttyGSO
Linux beaglebone 4.1.15-ti-rt-r43 #1 SMP PREEMPT RT Thu Jan 21 20:13:58 UTC 2016
armv71
The programs included with the Debian GNU/Linux system are free software;
the exact distribution terms for each program are described in the
individual files in /usr/share/doc/*/copyright.
Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
permitted by applicable law.
root@beaglebone:~# date
Fri Dec 2 09:47:02 UTC 2016
root@beaglebone:~# |
1 ° 1 B
Figure A4. PuTTY command code (continued)
& COM4 - PuTTY
Debian GNU/Linux 8 beaglebone ttyGSO
BeagleBoard.org Debian Image 2016-01-24
Support/FAQ: http://elinux.org/Beagleboard:BeagleBoneBlacfc_Debian
UAQHI support patrmccormack@gmail.com
The IP Address for usbO is: 192.163.7.2
beaglebone login: root
Last login: Fri Dec 2 09:46:59 UTC 2016 on ttyGSO
Linux beaglebone 4.1.15-ti-rt-r43 #1 SMP PREEMPT RT Thu Jan 21 20:13:58 UTC 2016
armv71
The programs included with the Debian GNU/Linux system are free software;
the exact distribution terms for each program are described in the
individual file3 in /usr/share/doc/*/copyright.
Debian GNU/Linux comes with ABSOLUTELY NO WARRANTY, to the extent
permitted by applicable law.
root@beaglebone:~# date
Tue Dec 6 11:30:24 UTC 2016
root@beaglebone: ~# []
-•
Figure A5. PuTTY command code (continued)
31

-------
Data were extracted and archived weekly without restarting the UNEP pod using the following
commands in the psftp console:
psftp - open
psftp mget -r data
The data were copied into the windows folder that stores the psftp.exe file. Data were archived in
a different folder to prevent other historical files from being amended during the data extraction
process.
The directory structure after the final extraction on 12/06/16 is shown in Figure A6.
File Edit View Tools Help
Organize* Include in library ~ Burn New folder	|EE ~ a #
-J Favorites
Name
Date modified
Type
H Desktop
A> Dayl
12/6/201611:29 AM
File folder
Downloads
i, Day2
12/6./201611:29 AM
File folder
Recent Places
ii Day3
12/6/201611:29 AM
File folder

. Day4
12/6/201611:29 AM
File folder
^ Libraries
it Day5
12/6/201611:29 AM
File folder
fj) Documents
.. Day6
12/6/201611:29 AM
File folder
Music
X Day7
12/6/201611:29 AM
File folder
B Pictures
Day8
12/6/201611:29 AM
File folder
3 Videos
i> Day9
12/6/201611:29 AM
File folder

... DaylO
12/6/201611:29 AM
File folder
Computer
i. Dayll
12/6/201611:29 AM
File folder

.. Dayl2
12/6/201611:29 AM
File folder
% Network
M Dayl3
12/6/201611:29 AM
File folder

Dayl4
12/6/201611:29 AM
File folder

i> Dayl5
12/6/201611:29 AM
File folder

J* Dayl6
12/6/201611:29 AM
File folder

i> Dayl7
12/6/201611:29 AM
File folder

J, Dayl8
12/6/201611:29 AM
File folder

ii Dayl9
12/6/201611:29 AM
File folder

U Day20
12/6/201611:29 AM
File folder

Ji Day21
12/6/201611:29 AM
File folder

U Day22
12/6/201611:29 AM
File folder

i, Day23
12/6/201611:29 AM
File folder

ii Day24
12/6/201611:29 AM
File folder

Day25
12/6/201611:29 AM
File folder

Day26
12/6/201611:29 AM
File folder

Day27
12/6/201611:29 AM
File folder

u Day28
12/6/201611:29 AM
File folder

.. Day29
12/6/201611:29 AM
File folder

A. Day30
12/6/201611:29 AM
File folder
Figure A6, Example file directory
The directory structure did not provide for separate folders for each month, resulting in new data
and data from the previous month (but same day of the month) being integrated together in the
same file.
Inside these folders, raw hourly data were stored in separate files, as shown below in Figure A7.
32

-------
File Edit View Tools Help
Organize ~ Include in library ~ Burn New folder	j§~ ~ Q] ©
,-Z Favorites
Name
Date modified
Type
Size
H Desktop
[£!] OPC_Data_hour_0.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
& Downloads
fiJj OPC_Data_hour_l.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
Recent Places
filf] 0PC_Data_hour_2.csv
12/6/201611:29 AM
Microsoft Excel C...
13 KB

fil] 0PC_Data_hour_3.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
^1 Libraries
fi|] OPC_Data_hour_4.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
H Documents
fia] 0PC_Data_hour_5.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
qfr Music
fil'1 OPC_Data_hour_6.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
E Pictures
fia| OPC_Data_hour_7.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
H Videos
fiaj OPC_Data_hour_8.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fial OPC_Data_hour_9.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
Computer
fij] OPC_Data_hour_10.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fi|i] OPC_Data_hour_ll.csv
12/6/2016 11:29 AM
Microsoft Excel C...
12 KB
% Network
A' OPC_Data_hour_12.c5v
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fill 0PC_Data_hour_13.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

t£a~] OPC_Data_hour_14.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fil] OPC_Data_hour_15.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fi£] 0PC_Data_hour_16.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fiii 0PC_Data_hour_17.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fiJ] OPC_Data_hour_18.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fia] OPC_Data_hour_19.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fill OPC_Data_hour_20.csv
12/6/201611:29 AM
Microsoft Excel C...
13 KB

© OPC_Data_hour_21.csv
12/6/201611:29 AM
Microsoft Excel C...
13 KB

fill OPC_Data_hour_22.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB

fil'l OPC_Data_hour_23.csv
12/6/201611:29 AM
Microsoft Excel C...
12 KB
Figure A7. Example hourly data file directory
Data for Different Months in the Same File
Below are two examples in which data from a previous month's data collection are mixed in with
data from the current month. Figure A8 shows that every other record is data from a previous
month. Figure A9 shows only two records that are data from a previous month. The September
data records are thought to be data previously collected by UNEP in testing the device that
remained in the processor's memory.
33

-------

A
8
C
D
E
F
G
H
1
J
K
L
1
Date
SysTime
Temp
RH
Dewpoin
PM1
PM2.5
PM10
SO2AIN0
S02AIN1
N02AIN2
N02AIN3
2
11/4/2016
0:00
25.01
61.09828
17.01777
327.5334
339.0214
339.2819
0.334687
0.341279
0.22478
0.223726
3
12/4/2016
0:00
13.16
43.13038
0.903038
42.19044
44.27616
45.51957
0.324053
0.338203
0.226802
0.222319
4
11/4/2016
0:01
24.95
61.30525
17.02604
65.873
72.86686
85.01807
0.334292
0.341894
0.226186
0.224165
5
12/4/2016
0:01
13.14
43.00455
0.863152
41.04337
43.12896
43.67744
0.324756
0.338027
0.227593
0.222363
6
11/4/2016
0:02
24.86
61.59757
17.00468
65.07874
72.48885
89.9003
0.33438
0.343125
0.226582
0.223154
7
12/4/2016
0:02
13.16
42.87984
0.812264
40.46441
42.524
44.49227
0.3248
0.338247
0.22645
0.2221
8
11/4/2016
0:03
24.87
61.59757
17.00468
64.61438
71.09325
83.72368
0.334204
0.341367
0.225659
0.223066
9
12/4/2016
0:03
13.12
42.9078
0.813018
41.92199
44.00334
45.63352
0.323701
0.335918
0.227153
0.221704
10
11/4/2016
0:04
24.86
61.59757
16.98502
64.66737
70.55445
78.01881
0.335303
0.342554
0.225703
0.223682
11
12/4/2016
0:04
13.12
42.93927
0.823202
40.95029
42.83891
43.51115
0.325415
0.340093
0.227593
0.221704
12
11/4/2016
0:05
24.84
61.68631
17.01056
65.65934
72.23067
86.68894
0.334512
0.340752
0.226582
0.223726
13
12/4/2016
0:05
13.12
43.06512
0.85371
39.75611
41.66798
42.81568
0.32647
0.339477
0.226802
0.223022
14
11/4/2016
0:06
24.83
61.77824
17.0341
66.14
72.62684
80.80582
0.334028
0.343301
0.226494
0.223682
15
12/4/2016
0:06
13.1
43.18738
0.856498
39.20398
40.97005
41.10087
0.324097
0.339082
0.227197
0.222231
16
11/4/2016
0:07
24.83
61.71695
17.00659
66.36286
72.77432
85.51762
0.334951
0.342114
0.225967
0.223813
17
12/4/2016
0:07
13.12
43.15948
0.874704
40.13837
42.27982
43.82177
0.326162
0.33895
0.22645
0.221836
18
11/4/2016
0:08
24.86
61.65888
17.03023
67.41539
74.00703
84.00356
0.33583
0.34185
0.227021
0.223594
19
12/4/2016
0:08
13.1
43.22
0.886179
41.20727
43.13165
44.69537
0.325283
0.33873
0.226494
0.222012
20
11/4/2016
0:09
24.87
61.65888
17.03023
67.05836
73.83238
82.07001
0.334468
0.341191
0.225923
0.223506
21
12/4/2016
0:09
13.11
43.18974
0.885511
41.79809
43.83794
44.35567
0.324228
0.337676
0.227109
0.2221
22
11/4/2016
0:10
24.84
61.77662
17.02227
66.27876
73.20253
85.45659
0.334556
0.341499
0.22645
0.224077
23
12/4/2016
0:10
13.1
43.15712
0.875389
39.59308
41.41007
41.91201
0.325547
0.338467
0.227153
0.221924
24
11/4/2016
0:11
24.8
61.98614
17.05733
66.57748
73.4444
84.45973
0.334204
0.34185
0.226186
0.223989
25
12/4/2016
0:11
13.1
43.21882
0.895626
40.3156
42.23076
44.47506
0.326206
0.339829
0.226186
0.222012

Figure A8. Example of suspected monthly data file reporting error
215
11/8/2016
3:33
8.82
60.30468
1.593058
0.001565
0.117049
0.412528
0.330249
0.33939
0.226011
0.222275
44.249
63.39
-49.989
-55.989
-4
1.0354
67.21979
216
11/8/2016
3:34
8.93
59.95524
1.620155
0.002529
0.112996
0.232267
0.330864
0.339785
0.225176
0.221792
44.864
63.785
-50.824
-56.824
-4
1.0321
67.61479
217
11/8/2016
3:35
8.83
60.49935
1.691278
0.002829
0.173538
0.528631
0.330908
0.339521
0.225308
0.221704
44.908
63.521
-50.692
-56.692
-4
1.0351
67.35079
218
11/8/2016
3:36
8.87
60.35985
1.62692
0.0003
0.060547
0.296379
0.331787
0.339477
0.224736
0.222451
45.787
63.477
-51.264
-57.264
-4
1.0339
67.30679
219
11/8/2016
3:37
8.82
60.38248
1.593058
0
0
0
0.329546
0.339346
0.225
0.222539
43.546
63.346
-51
-57
-4
1.0354
67.17579
220
11/8/2016
3:38
8.79
60.22974
1.540963
0.002829
0.173557
0.52869
0.330117
0.339961
0.225132
0.221924
44.117
63.961
-50.868
-56.868
-4
1.0363
67.79079
221
11/8/2016
3:39
8.8
60.64536
1.588297
0.0003
0.060546
0.296377
0.331611
0.340181
0.225615
0.221484
45.611
64.181
-50.385
-56.385
-4
1.036
68.01079
222
9/8/2016
3:40
29.94
42.21175
15.70961
430.1873
453.0135
465.6525
0.354375
0.355649
0.221748
0.219463
68.375
79.649
-54.252
-60.252
19.88
0.4012
59.59879
223
224
11/8/2016
3:40
8.86
60.00231
1.607291
0.001264
0.056495
0.116127
0.331919
0.339521
0.225615
0.222671
45.919
63.521
-50.385
-56.385
-4
1.0342
67.35079
9/8/2016
3:41
30.22
41.8752
15.81527
101.8284
112.3463
132.1761
0.347519
0.351474
0.22188
0.220825
61.519
75.474
-54.12
-60.12
22.64
0.3956
52.66379
225
11/8/2016
3:41
8.8
60.35985
1.567362
0.000368
0.105904
0.897723
0.331348
0.338423
0.224912
0.222539
45.348
62.423
-51.088
-57.088
-4
1.036
66.25279
226
11/8/2016
3:42
8.77
60.49604
1.613386
0.001933
0.222971
1.31038
0.329238
0.339873
0.225088
0.221792
43.238
63.873
-50.912
-56.912
-4
1.0369
67.70279
227
11/8/2016
3:43
8.79
60.25898
1.582724
0
0
0
0.329502
0.339038
0.224517
0.221528
43.502
63.038
-51.483
-57.483
-4
1.0363
66.86779
228
11/8/2016
3:44
8.74
60.03158
1.602361
0
0
0
0.329722
0.338467
0.225044
0.222803
43.722
62.467
-50.956
-56.956
-4
1.0378
66.29679
229
11/8/2016
3:45
8.78
60.18113
1.447417
0.001563
0.116947
0.412171
0.329634
0.338994
0.224648
0.222495
43.634
62.994
-51.352
-57.352
-4
1.0366
66.82379
Figure A9. Example of suspected monthly data file reporting error (continued)
Sensor Pod PM Sensor Spikes
As shown in Figures A10 and A11, anomalous spikes in PM2.5 and PM10 concentrations occurred
at the top of every hour (XX:00). These data were removed from the colocation correlation
analysis.
34

-------
57
11/4/2016
0:55
24.46
64.03549
74.7328 96.73585
58
11/4/2016
0:56
24.44
64.27158
74.11503 92.42737
59
11/4/2016
0:57
24.41
64.42154
74.59512 99.87106
60
11/4/2016
0:58
24.38
64.47373
76.1834 89.75808
61
11/4/2016
0:59
24.4
64.32382
74.45122 84.61952
62
11/4/2016
1:00
24.48
63.88361
324.6425 330.8294
63
11/4/2016
1:01
24.42
64.11649
73.111 83.18977
64
11/4/2016
1:02
24.38
64.32382
74.49226 97.53886
65
11/4/2016
1:03
24.38
64.32382
70.95723 73.54578
66
11/4/2016
1:04
24.38
64.38278
76.88055 89.86658
67
11/4/2016
1:05
24.38
64.35415
71.20496 89.90846
Figure A10. Anomalous concentration spikes
115
11/4/2016
1:53
24.02
65.68365
82.99122
119.5055
116
11/4/2016
1:54
24.06
65.56647
82.55427
109.7412
117
11/4/2016
1:55
24
65.8309
80.70968
103.0655
118
11/4/2016
1:56
24.01
65.68021
78.60008
91.88626
119
11/4/2016
1:57
24
65.74049
77.47578
96.93566
120
11/4/2016
1:58
23.99
65.79904
78.03426
98.67557
121
11/4/2016
1:59
24.05
65.56647
75.6331
85.33802
122
11/4/2016
2:00
24.04
65.68365
340.7785
343.86851
123
11/4/2016
2:01
24.1
65.36554
76.55142
105.2826
124
11/4/2016
2:02
24.04
65.56819
77.77175
88.68311
125
11/4/2016
2:03
24.09
65.45266
76.60575
83.38798
126
11/4/2016
2:04
24.1
65.27667
78.25141
92.45849
127
11/4/2016
2:05
24.1
65.30515
76.72618
90.39031
128
11/4/2016
2:06
24.13
65.10053
77.76107
88.47563
129
11/4/2016
2:07
24.15
65.04008
73.24411
84.04717
130
11/4/2016
2:08
24.06
65.4794
76.91001
85.59264
Figure A11. Anomalous concentration spikes (continued)
Repetition of Time Stamp
Same time stamps:
28762
11/23/2016
23:58
14.02
40.67147
0.142399
0.433776
28763
11/23/2016
23:59
13.96
41.36422
0.23S44
0.772264
28764
11/24/2016
0:00
13.88
41.92297
0
0
28765
11/24/2016 0:02
13.87
42.1737
0.096044
0.338497
28766
11/24/2016
0:02
13.86
42.36333
0.192078
0.67696
28767
11/24/2016 0:03
13.87
42.33404
0.099361
0.486379
28768
11/24/2016 0:05
13.86
42.33288
0.096035
0.338467
28769
11/24/2016 0:06
13.87
42.13978
0.145707
0.581618
28770
11/24/2016 0:07
13.86
41.82689
0.096059
0.338552
Figure A12. Repetitious time stamps
In Figure A12, note that the PM data for identical time stamps (highlighted) are different. For
these analyses, the second instance of the same time stamp was deleted and not included in the
35

-------
correlation analysis. Though it is likely that the first instance shown in this illustration describes
time stamp 0:01 and the second describes for timestamp 0:02, determining which data record to
delete is inconsequential since a full set of data records (five 1-minute measurements) was
required to compute 5-minute averages based on the 90% data completeness rule.
Investigation of the NO2 Concentration Computation
A macro file provided by UNEP (Gas-ppb.xlsm) was used to compute the gas concentrations
from the electronic output of the sensor. All the reported NO2 concentrations in the resulting
processed data files were small negative numbers. A cursory look at these results would lead one
to suspect that the sensor was not functioning properly. However, these concentrations are based
on a temperature correction algorithm and other correction factors and applied to the raw sensor
output through the macro code. Such a situation creates numerous opportunities for errors to be
introduced, so the algorithm, correction factors, and their application in the macro were
examined in detail. The findings are best illustrated through an example calculation performed
manually.
Alphasense Ltd., the manufacturer of the gas sensors used in the UNEP pod, has developed
correction algorithms for correcting zero background currents due to temperature changes. These
are provided by the Alphasense Application Note AAN 803-02, "Correcting zero background
currents of four electrode toxic gas sensors due to temperature changes". Both the recommended
algorithm and the alternate algorithm for NO2 provided in this application note are for the "A"
series sensors. However, the UNEP pod uses the "B" series sensor for NO2. The implications of
applying the algorithms to the B series sensor are not discussed in the application note.
An example calculation was performed manually and compared with the results provided by the
macro to confirm the macro code or identify errors. Two errors affecting NO2 computations were
found and corrected (discussed under the heading "Macro Code Errors" later in this Appendix).
All computations going forward were based on the corrected macro. The following explanation
demonstrates an example calculation for data record 11/18/2016, 13:00 EST, which was selected
based on the high NO2 concentrations reported on that day by the reference monitors to ensure a
robust measurement.
For NO2, the recommended algorithm is WEc = WEt - m*AET, where WEc is the corrected
working electrode, WEt is the uncorrected working electrode, ni is the temperature dependent
correction factor, and AEt is the uncorrected auxiliary electrode.
Alphasense supplies electronic offsets that are unique to each sensor. For the NO2 sensor in the
UNEP pod, these offsets (as provided to the US EPA by UNEP) were 282 mV for the working
electrode and 276 mV auxiliary electrode. These electronic offsets must be subtracted from the
36

-------
raw readings before applying the temperature corrections. The raw WE and AE are reported in
units of volts (V), so these values must first be converted to mV.
The following is an example of the calculation for data record 11/18/2016, 13:00 EST, which
was selected based on the high NO2 concentrations reported on that day by the reference
monitors, ensuring a robust measurement.
Raw WE = 0.224165 V = 224.165 mV
Raw AE = 0.220737 V = 220.737 mV
Raw WE minus offset = 224.165 - 282 = -57.835 mV = WEt (uncorrected working
electrode, EC2_WE_mV, column P)
Raw AE minus offset = 220.737 - 276 = -55.263 mV = AEt (uncorrected auxiliary
electrode, EC2_AE_mV, column O)
A temperature of 31.26 degrees C was reported by the UNEP pod for this data record. The
temperature dependent correction factor m was 0.3728, as interpolated from discrete values
supplied in table in the Alphasense Application Note. (This computation was also confirmed
manually.)
These results were applied in the recommended algorithm as follows:
WEC = WEt - nT*AET = -57.835 mV - 0.3748*(-55.263) = -37.122 mV
The algorithm requires each term of the equation to be in units of nA rather than mV, so this
result must be converted to nA using the correction factor -0.73 nA/mV (supplied by Alphasense
and provided to the US EPA by UNEP):
WEC (nA) = (-0.73 nA/mV)*(-37.122 mV) = 27.099 nA
To obtain the concentration units, this result is divided by the sensitivity factor (SF) -380.5
nA/ppm (supplied by Alphasense and provided to the US EPA by UNEP):
NO2 concentration = WEc (nA)/SF = 27.099 nA/(-380.5 nA/ppm) = -0.0712196 ppm
The concentration is converted to ppb by multiplying by 1000, yielding a concentration of -71.2
ppb. The concentration reported by the reference monitor for this time period was 55.5 ppb. But
for the signage issue, this would be a reasonable comparison. A cursory review of the output for
some other data records comparing the macro-computed NO2 concentrations with the reference
monitor NO2 concentrations shows that this observation is backed by reasonable evidence, and
that it is not unreasonable to pursue this line of inquiry to solve the NO2 measurement problem.
It is recommended that such a line of inquiry involve detailed sharing and review of these
findings with the manufacturer.
37

-------
The Effect of Reported Temperature on Gas Concentration Computations
As reported in section 3.2, the temperature sensor in the UNEP pod showed a positive bias > 7°C
compared with the temperatures reported by reference devices. It was hypothesized that this bias
might be the result of heating within the encasement of the UNEP pod, or of a calibration bias.
Either hypothesis raises the question of which temperature should be used in the gas concentration
algorithms. If the true temperature in the UNEP pod that is influencing the gas sensor measurements
is lower than what is being reported, then what is the effect on the computed concentration?
To resolve this issue, the ambient reference temperature of 22.6 °C was substituted for the UNEP
pod temperature of 31.26 °C for the above example. The corresponding temperature dependent
correction factor is then 0.548. Applying this to the example yields an NO2 concentration of -
52.857 ppb, compared with the reference concentration of 55.5 ppb. This would be quite a good
comparison were it not for the signage issue.
Macro Code Errors
Errors in the macro code were discovered during the investigation of the computation of the NO2
concentrations. The errors, and their recommended solutions, are summarized here:
1)	In computing the uncorrected working electrode value (column P - EC2_WE_mV), the
code referred to the column for the raw auxiliary electrode (column K - N02A1N2)
rather than to the column for the raw working electrode (column L - N02A1N3). The
following is the code affected:
•	Change Range ("P") select
•	ActiveCell.Formula R1C1 = "(RC[-5]xl000]-282
The US EPA recommended that the relative cell indicator be change from RC[-5] to RC[-
4] to refer to the correct column, and UNEP supplied the corrected code.
2)	Sensitivity factors are in units of nA/ppm, but the macro-produced spreadsheet identified
computed concentrations as ppb. The US EPA recommended that a factor of 1000 be
applied in the code to report the results as ppb, and UNEP supplied the corrected code.
Through further examination of the data files, a data sorting error was discovered. The date/time
stamps in the processed output showed that some time periods had apparently been omitted.
Upon closer examination, it was found that a block of records that should have appeared earlier
in the time column were added instead to the end of the column after 23:59. The issue was found
to lie in the sorting subroutine, wherein all sorting code referred to rows 2 through 1440 (or less).
The US EPA recommended that the sorting code be amended to refer to rows 2 through 1441
(assuming no extra records are included from previous months). The US EPA determined that
the best approach to managing these data files, some of which contained extra data records, was
to sort the data again manually following application of the macro to complete the data sorting
and put records in the proper order.
38

-------
PRESORTED STANDARD
United States
Environmental Protection
Agency
Office of Research and Development (8101R)
Washington, DC 20460
Official Business Penalty
for Private Use $300
POSTAGE & FEES PAID
EPA
PERMIT NO. G-35
oEPA

-------