How to Evaluate Low-Cost Sensors

by Collocation with Federal

Reference Method Monitors

National Exposure Research Laboratory

Office of Research and Development

1
 Contributors

EPA - Office of Research and Development

Teri Conner, [email protected]
Andrea Clements, [email protected]
Ronald Williams, [email protected]

EPA - Office of Air Quality Planning and Standards

Amanda Kaufman, [email protected]

2
 Introduction and Objective
Recent technological advances in air quality monitoring are providing
smaller, lower-cost, sensor-based monitors.

• These low-cost sensors can measure some of the same air pollutants that
more costly reference monitors can measure.
• But how good are they?
The goal of this presentation is to provide instructions for evaluating low-cost
sensors by collocating them with reference monitors and comparing
measurements of the two to determine the “trustworthiness” of low-cost sensors.
3
 Background: Reference Monitors

What are Reference Monitors? Why do we need them?

• The Clean Air Act requires every state to establish a network of air monitoring stations
to monitor criteria air pollutants that must meet National Ambient Air Quality
Standards (NAAQS). Not all criteria pollutants are measured at all stations.
• The measurement instruments and methods used at these regulatory monitoring
stations are called Federal Reference Methods (FRM) or Federal Equivalent Methods
(FEM).

• The FRM/FEM instruments must meet rigorous standards for accuracy and reliability
(see 40 CFR part 53 for details). These instruments are “The Gold Standard”.
• In this presentation, these FRM/FEM instruments will be referred to as reference
monitors for simplicity.
4
 Background: Reference Monitors vs
Low-Cost Sensors
Here is how low-cost sensors compare with reference monitors:
Reference Monitors Low-Cost Sensors
Typical Purchase Cost Range $15,000 to $50,000 $100 to $2500
Operating Expense Expensive Inexpensive
Fixed Location Portable
Siting Location
(building/trailer needed) (basic weather shielding)
Staff Training Highly trained technical staff Little or No Training
Unknown and may vary
Known and consistent quality from sensor to sensor and
Data Quality
in a variety of conditions in different weather
conditions
10+ Years Short (1 year) or Unknown
Operating Lifetime (calibrated and operated to (may become less
maintain accuracy) sensitive over time)
Used for Regulatory Monitoring? Yes No

5
 Background: Low-Cost Sensors

How can low-cost sensors be used?

• The low-cost and portability of
these sensors makes it possible
to measure air quality in more
locations.
• Many of these sensors can
report measurements at a
frequency on the order of
minutes, or even seconds,
making it possible to learn Orange dots are sensor monitoring locations in Newark area.
about changes in air quality
throughout the day.

Low-cost sensors may be used to locate pollution hotspots, identify sources

of pollution, supplement fixed-site monitoring data, measure personal
exposure to pollutants, educate, and enhance air quality awareness.

6
 Background: Low-Cost Sensors

The low cost, portability, and ease of use of sensor technology make it possible
for citizens to participate in air quality monitoring within their communities.
Here are some examples:
• Ironbound Community, Newark, New Jersey
• DISUR (Desarrollo Integral del Sur), Ponce, Puerto Rico
• Federal Crowdsourcing and Citizen Science Toolkit

Citizens learning to operate

EPA and citizens at a monitoring site. EPA sensor pods.
7
 Background: Low-Cost Sensors

These low-cost sensors

sound great! But how good are they?

8
 Evaluating Low-Cost Sensors:
Collocation
What is collocation?
• Collocation is the process by which a reference monitor (FRM/FEM) and
non-reference monitor (sensor) are operated at the same time and place
under real world conditions for a defined evaluation period.
• Sensor performance can be evaluated and data accuracy improved by
comparing sensor data with reference monitor data.
• Here are some examples of sensor evaluations by collocation with reference
monitors:
o Particulate Matter Sensor Evaluation Report
o Sensor Evaluation Report

More examples of collocation can be found

on the Air Sensor Toolbox web page.
9
 Purpose

• The materials provided in this presentation will give you an understanding of

the important considerations for any collocation effort.
• The goal of a collocation effort is to evaluate sensor data quality and help
answer some important questions like these:

Are my sensor
Why are my two sensors readings accurate?
saying different things? Is this number bad?
What does it mean?

Use collocation to help you

answer these questions!

10
 Approach

Collocation is a multi-step process.

You can link to each step by clicking in this graphic:
Step 5: Using
Sensors
Step 4: Data
Step 3: Data Comparison
Recovery and
Review
Step 2: Making
Measurements
Step 1:
Planning

11
 Step 1: Planning a Sensor Evaluation

What sensor are you evaluating?

• If you are planning to evaluate a sensor, you
probably have already determined what
pollutant you want to measure, and what
sensor you want to use to measure it. Keep
in mind that not all criteria (regulated)
pollutants can be measured with low-cost
sensors.
• If not, start by asking a question to get you
started – something like “What time of day
are particulate matter concentrations highest
in my neighborhood?” Think about logistics –
the who, what, where, when, why, how.

The Air Sensor Guidebook is a good place to learn about air quality, pollutants,
and their sources (chapter 2) and what to look for in a sensor (chapter 3).

12
 Step 1: Planning a Sensor Evaluation

How many sensors do I need?

• If possible, acquire at least 3 units of the same sensor.
Why should I evaluate three (or more!) sensors?
• It will be easier to recognize a faulty sensor.
• You’ll have more data to evaluate the accuracy and variability of the sensor
measurement.

Example of sensors being tested in triplicate 13

 Step 1: Planning a Sensor Evaluation
Besides sensors, what equipment and supplies do I need?
• What power source do I need?
o AC power: Make sure you have sufficient numbers and lengths of power cords, and a place
to plug everything in.
o Other: Solar panels/batteries/charging cables
• How is the data stored or transmitted?
o Storage media: Typically micro SD cards
o Cellular transmission: Is there sufficient cellular signal or WiFi? Where is the data
transmitted? How is it transmitted? Do I need to buy a dedicated cellular plan?
o Computer connection: Cords to connect sensor to computer (if required)
• What computer and software do I need?
o Sensor-specific software: Provided by sensor manufacturer
o Other software: For data analysis (e.g., Excel), data download and transmission (e.g., PuTTY)
o Computer: Must be sufficient to support all hardware and software

Before you start, read the user guide or instructions provided with your
sensor. It’s a good idea to test everything before you do your field evaluation
to better understand your sensor and identify any potential problems.
14
 Step 2: Making the Measurements

Do I need anything else?

• Yes! An important component of any field study or
measurement project is a project notebook.
• It should be a bound notebook, and weather-proof if used
outdoors for field measurements.
• Mark out errors by drawing a single line through the error,
and initial the change.
• Use the project notebook to document your activities in the field, and well as
before and after field deployment.
• Here are some examples of things to write in the project notebook:
o Tests and observations before the field evaluation
o Details of field site visits – time, date, and duration; your activities and
observations; status of sensors; nearby activities and unusual weather that
may influence the measurements; problems observed and solutions
o Interesting observations and decisions about data after deployment during
data analysis.
15
 Step 1: Planning a Sensor Evaluation

I’ve selected my sensor. Now what?

• One option is to contact your local air quality/natural resource board or agency.
A good place to start is the Association of Air Pollution Control Agencies or the
National Association for Clean Air Agencies.
• Reaching out to these agencies before starting a project can help you define your
projects and goals, and direct you to the right contacts. They can also help you
identify locations of reference monitoring sites, since they usually run them.
• Be aware that access to reference monitoring sites used for regulatory purposes
will likely be restricted. You would have to gain permission from the local air
quality agency to conduct research in close proximity to their secure monitoring
locations.
The Air Quality System (AQS) contains
ambient air pollution data collected by
EPA, state, local, and tribal air pollution
control agencies from over thousands of
monitors. Here you can find instructions
and links for obtaining AQS data.
16
 Step 1: Planning a Sensor Evaluation

What other options do I have for collocating with a reference monitor?

• The reference (FRM/FEM) monitor used for collocation comparison does not need
to be located at an official regulatory site.
• But it does need to be operated following approved quality assurance protocols.
This can be done by any knowledgeable and reputable entity such as the following:
o Federal, tribal, state, or local air quality/natural resources agency
o University partner – typically with expertise in environmental engineering,
atmospheric sciences, and similar fields – your local agency may be able to
recommend university contacts
o Air quality contractor

To help prepare for these

initial conversations, it is
recommended to review the
questions listed in Appendix A
of the Air Sensor Guidebook.
17
 Step 1: Planning a Sensor Evaluation
How do I plan to collect data for scientific evaluation?
• Write up a plan (called a Quality Assurance Project Plan) that answers these
questions:
o What question are you trying to answer? Click here to go back to
the Approach steps slide.
o What data are needed?
o Who will use the data and how will it be used?
o Are there specific requirements for the data? Time of day? Season?
Measurement duration?
o How will data quality be evaluated?
View this video to learn more:
o Collocation with reference data! Quality Assurance Training Video
o Who will do the work?
 List participants and their responsibilities in the project.
 Include participants who can solve technical problems and understand data.

Download a template for your plan:

Quality Assurance Project Plan for Citizen Science Projects 18
 Step 2: Making the Measurements

Where should I set up my sensors?

• Ideally, sensors and reference monitors should be within 10 meters (30 feet)
of each other, and the inlets (where air enters the sensor/monitor) should be
at about the same height.
• The movement of air around the sensor should not be blocked or hindered by
any other structure or device.
• They should be protected from the weather.

Reference Monitor Inlet

Sensor Inlet

Example of a custom-made weather

Example of sensor inlet located shelter for sensor evaluation.
near reference monitor inlet.
19
 Step 2: Making the Measurements

How do I make a useful comparison of the sensor and reference data?

• Find out how frequently the reference monitor reports data. Is it every 5 minutes?
Every hour? Every 24 hours (typical of filter-based measurements)?
• Can your sensor report data at least as frequently, and at the same times, as the
reference monitor? If possible, choose a sensor data reporting frequency somewhat
shorter than the reference monitor data reporting frequency.

Here’s an example of how to match sensor and reference data

measurement periods:
1. Situation: The reference monitor reports data every 5 minutes. The
sensor data frequency choices are every 5 minutes, 1 minute, or 1 second.
2. Choose sensor data frequency: Choose 1 minute. (Choosing 1 second
creates more data points than I need.)
3. Average sensor data: Take every five 1-minute sensor data points and
average them to get a 5-minute data point to match the reference
monitor’s data reporting times.

20
 Step 2: Making the Measurements

How long should I How many data

collect data? points do I need?

Here are some suggestions for data collected for different time intervals:
• 24-hour data – 1 month of successful data collection - ~30 valid data pairs
• 1-hour data – 2 weeks of successful data collection - ~500+ valid data pairs
• 5 min data – 1 week of successful data collection - ~2000+ valid data pairs

21
 Step 2: Making the Measurements

Other factors to consider:

• Many pollutants follow seasonal patterns. If possible, do
collocation measurements when the pollutant of interest
is more likely to be at higher concentrations and more
likely to be changing throughout the measurement period.

• Extreme temperature and relative

humidity can affect sensor readings.

• Some pollutants or conditions may interfere with the

pollutant you are trying to measure. Refer to the
specifications provided by your sensor manufacturer
for more information.

• Some sensors, like those for gases, may gradually lose their responsiveness over
time and need to be replaced, typically after one year. This happens whether
they are used or not. How can you tell? Their measurements will be trending
downward when the reference measurements are not.

22
 Step 2: Making the Measurements

Troubleshooting:
• Making measurements of any kind can have its challenges,
especially in outdoor environments.
• Make sure you understand your sensor and what normal
function looks like.
• Visit your monitoring site on a regular basis to catch
problems and fix them before too many data points are lost.
• There are many things that can go wrong, and not all problems are due to a
malfunctioning sensor. Here are a few common issues to check:
o Is the sensor getting power? Check your power source or breaker.
o Has there been an unusual weather event?
o Is there evidence of animals or insects interfering with the measurement?
o Are all cable connections secure?
o Is the inlet or outlet of the sensor blocked?
Click here to go back to
o NOTE ALL OBSERVATIONS IN YOUR PROJECT NOTEBOOK! the Approach steps slide.

23
 Step 3: Data Recovery and Review
Now that I have data, what should I do?
• Once you have data, it’s important to review it to evaluate quality and identify
problems.
• It’s a good idea to review your data during the collection phase to identify problems
that may affect your data and correct them, as well as at the end of your data
collection to evaluate the whole data set.
• Things to look for when reviewing your data:
o Abnormally high or low values (outliers)
o Expected Patterns
o Interferences
o Drift or Shift

These features are defined, with examples, on the following slides.

Use information and observations recorded in your field notebook to
help you understand any problems you find in the data, and to help
you decide how to handle them (e.g., exclude certain data points.)
24
 Step 3: Data Recovery and Review

Outliers – data points that look out of place –

much lower or higher than nearby data points

Field notes confirm I

was visiting the sensor
and left my car idling.
That likely explains
these high values.

This zero value is unusual. I wonder

if this is a problem with the sensor
or maybe data communication. I
will keep an eye on the data to see if
it happens again.

25
 Step 3: Data Recovery and Review

Expected Patterns – could be seasonal, day/night, or weekday/weekend

patterns - absence of expected patterns may indicate a problem with
your sensor or with your measurement approach

Ozone concentrations
peak in late afternoon,
as expected.

26
 Step 3: Data Recovery and Review

Interferences - may have a positive or negative effect on sensor response - can

include pollutants or other chemical compounds that are not of interest, weather
conditions, dirt/dust/insects.

The sensor is supposed to be

measuring NO2, but it’s reporting
concentrations higher than the
combined reference NO2 and O3
concentrations. Could be a
measurement interference.

27
 Step 3: Data Recovery and Review

Drift or Shift - a gradual (drift) or sudden (shift) change in a sensor’s response

characteristics over time – can be positive or negative - may lead you to wrongly
conclude that concentrations have increased or decreased over time

Concentrations reported by the

sensor are increasing gradually
over time. Looks like the sensors
response is drifting.

Click here to go back to

the Approach steps slide.

28
 Step 4: Comparing the Data

Help! What do I do with all this data I have collected?

• Working with data sets and making comparisons may seem
challenging, but there are tools available to make the job easier.

Wrestling with data? Take it to the MAT!

• The MAT (Macro Analysis Tool) is an Excel-based user-friendly macro tool
developed by the EPA specifically for the purpose of comparing sensor data with
reference data.
• Data sets don’t have matching time-date stamps? Sensor data collected on a
different time interval than the reference data? Periods of missing data? No
problem – MAT takes care of that, and more!

29
 Step 4: Comparing the Data
Do I have to use the MAT (macro analysis tool) to compare my data?
• No, you can use any software or data analysis tool to compare your data, based on
your experience and comfort-level. But the MAT makes data comparison a lot easier!
Great! How do I use the MAT?
• The user inserts a column of sensor data onto the sensor page of the tool, and a
corresponding column of reference data onto the reference page of the tool.
• Next, the user enters information in the control panel on the setup page, including
the following:
o Pollutant name
o Start and stop dates and times
o Reported units and sampling intervals
o Desired averaging time interval
o Data completeness criteria
• Then press RUN – the MAT takes care of the rest!
A complete list of features and instructions is available on the first
page of the MAT. Click here to get a copy. 30
 Step 4: Comparing the Data

Are data values missing or unusable?

• No data collection effort is perfect!
• Some expected data values may be missing due to downtime of the measurement
device (e.g., power outages, maintenance).
• Some acquired data values may be unusable (e.g., nearby interfering activities, unusual
conditions), or outliers (data value that is abnormally different from nearby data
values). You may decide to exclude these from your calculations.
• Document your decisions about the data – what you did and why – in your project
notebook.

The MAT will allow you to flag these unusable

data points for exclusion from your analysis.

31
 Step 4: Comparing the Data
What is Data Completeness?
Data completeness refers to the amount of usable data that was obtained,
compared to the total amount expected. It is expressed as a percentage.
Click here to go back to the
Example: MAT overview slide.

• Suppose you want to take hourly data and average it to 24 hours.

• Now let’s say that one of the 24-hour periods in your data had only 18 out of 24 hours
of usable data.
• Now suppose that you decide you need each 24-hour period to be 90% complete. (Data
completeness is another quantity that can be selected in the MAT.)
• In this example, there are not enough usable data points to meet this requirement,
since the data is only 75% complete (18/24*100=75%), so an average would not be
reported for that 24-hour period.
• The more data points available to average (higher data completeness percentage),
the better that average represents the whole measurement period. Reporting the
data completeness percentage applied to the data helps you and others understand
your data better. 32
 Step 4: Comparing the Data

What information does the MAT produce?

Output page: Shows the two data sets being compared according to selections made in
the control panel. Here’s an example:
Averaged Date Time Reference Reference O3 Invalid Data Points Sensor O3 Invalid Data Points
10/13/2016 12:00 47.53 0 18
10/13/2016 13:00 51.58 0 59.32 0
10/13/2016 14:00 54.93 0 59.67 0
Date/time stamps for 10/13/2016 15:00 58.65 0 64.73 0
reference monitor and 10/13/2016 16:00 54.33 0 60.55 0 Here are the flagged
sensor have been matched. 10/13/2016 17:00 53.69 0 60.56 5 or missing data points.
10/13/2016 18:00 43.13 0 48.78 0
10/13/2016 19:00 29.21 0 33.57 0
10/13/2016 20:00 11.68 0 14.17 0
10/13/2016 21:00 6.08 0 7.72 0
Graph page: Shows a time series comparison of the two data sets, a correlation
10/13/2016 22:00
10/13/2016 23:00
7.42
26.40
0
0
8.42
30.90
0
0
comparison of the two data sets, plus some statistics.

This is information you typically need for your comparison no matter which
software tool you choose for your data analysis. The following slides show
examples and explain how to interpret the graphs and statistics.

33
 Step 4: Comparing the Data
One way to compare the sensor data with the reference data is to plot the data in
a correlation graph, as shown here:
y = mx + b

coefficient of determination
slope-intercept line

The line going through the data is called the “slope-intercept” line and is represented by the
equation y = mx + b. This equation is a statistical means of comparing the sensor data with
the reference data. The coefficient of determination, represented by R2, is a measure of how
close the data are to the slope-intercept line.
More information is provided on the following slides.
Here’s a great instructional video on the topic: y=mx+b video
34
 Step 4 : Comparing the Data

More about the coefficient of determination, R-squared

R-squared (R2 ) is a statistical measure of how close the data are to the slope-intercept line -
in other words, how much scatter is in the data. Its values range from 0 to 1. The closer R2
is to 1, the stronger the agreement between the sensor and the reference data.
R2 = 0.99 shows more consistent
agreement (less scatter) between sensor
and reference than R2 = 0.70

Click here to go back to the

correlation graph slide.

35
 Step 4 : Comparing the Data
More about the “slope-intercept” equation y = mx + b
• This equation represents the average behavior of the sensor data (vertical axis,
represented by y) compared with the reference data (horizontal axis, represented by x).
• The slope (m) shows how similar, or different, the sensor measurements are compared
with the reference measurements, on average. The farther “m” is from 1, the more the
sensor over- or under-reports the concentration. The closer “m” is to 1, the more your
sensor responds like (in proportion to) the reference monitor. The slope can be either a
positive or negative value.

Sensor over-reports
Sensor response close to reference

Sensor under-reports

Click here to go back to the

correlation graph slide.
36
 Step 4 : Comparing the Data

More about the “slope-intercept” equation y = mx + b

The intercept (b) shows what the sensor measurement will be, on average, when the
reference monitor is measuring zero concentration. The farther “b” is from zero, the more
the sensor over- or under-reports a concentration. The closer “b” is to zero on the y-axis
(vertical axis), the less the amount of “bias” in the sensor response.

more bias

Click here to go back to the Click here to go back to

correlation graph slide. the Approach steps slide.

less bias

37
 Step 5: Using Sensors

Suppose you want to take your sensors and do measurements away from a
reference monitor. How can you still make sure your sensor data represent
concentrations close to the true (reference) value?
• Remember, the slope-intercept equation y = mx + b relates the sensor measurements
to reference measurements:
𝐬𝐞𝐧𝐬𝐨𝐫 = 𝐦 ⋅ 𝐫𝐞𝐟𝐞𝐫𝐞𝐧𝐜𝐞 + 𝐛
• Use the values for m (slope) and b (intercept) you computed from the collocation
data and apply these to the new sensor data so it is in better agreement with the
reference data.
𝐦𝐞𝐚𝐬𝐮𝐫𝐞𝐝 𝐬𝐞𝐧𝐬𝐨𝐫 𝐜𝐨𝐧𝐜𝐞𝐧𝐭𝐫𝐚𝐭𝐢𝐨𝐧 − 𝐛
𝐚𝐝𝐣𝐮𝐬𝐭𝐞𝐝 𝐬𝐞𝐧𝐬𝐨𝐫 𝐜𝐨𝐧𝐜𝐞𝐧𝐭𝐫𝐚𝐭𝐢𝐨𝐧 =
𝐦
• Do this for each sensor you evaluate.
• Keep the R2 value in mind. The more scatter in the data (the farther R2 is from 1), the
less useful this adjustment will be.

38
 Step 5: Using Sensors

Here are some examples of how to adjust sensor data to be closer to

reference data
Example #1: Take a simple example where the b is zero, and m is 1.1.
 The equation becomes SENSOR = 1.1*REFERENCE.
 To make the new sensor values closer to what reference values would be,
you have to divide the measured sensor value by the slope 1.1.
Example #2: Now let’s make it more interesting. Suppose that b is equal to 1. That
means the sensor measures 1 unit higher than the reference.
 The equation then becomes SENSOR = 1.1*REFERENCE + 1.
 To do the right adjustment, first subtract b (=1 in this example), then divide
by m (=1.1 in this example).

In summary, the equation becomes

𝐌𝐞𝐚𝐬𝐮𝐫𝐞𝐝 𝐬𝐞𝐧𝐬𝐨𝐫 𝐜𝐨𝐧𝐜𝐞𝐧𝐭𝐫𝐚𝐭𝐢𝐨𝐧−𝐛
𝐀𝐝𝐣𝐮𝐬𝐭𝐞𝐝 𝐒𝐄𝐍𝐒𝐎𝐑 𝐜𝐨𝐧𝐜𝐞𝐧𝐭𝐫𝐚𝐭𝐢𝐨𝐧 =
𝐦
39
 Step 5: Using Sensors

Now that I’ve done my collocation data comparisons, how do I decide if my

sensors produce data that is good enough for my purposes?
• Depends on the purpose!
• Collocation results provide an understanding about key performance
characteristics of your sensor
• Go to the Air Sensor Guidebook to learn more. Chapter 4 provides information
on how to collect useful data. Chapter 5 describes different types of sensor
applications and performance goals for each.

Click here to go back to

the Approach steps slide.

40
 Additional Data Quality Information
Accuracy, Bias, and Precision – Important data quality parameters
• Accuracy is how close a measured value is to the actual (true) value.
• Bias is a systematic (built-in) error which makes all measurements wrong by a certain
amount.
• Precision is how close the measured values are to each other.
• These are illustrated in the following diagram:

More information is provided on the following slides.

Here’s a link to some great instructional videos: Bias and Precision Videos
41
 Additional Data Quality Information

More about bias

Bias is a persistent error in a measurement process that causes all measured values to
be too high or too low by a certain amount, compared to the true (reference) value. This
is the intercept value in the slope-intercept equation y = mx + b.

42
 Additional Data Quality Information

More about precision

Precision is a measure of how close repeated measurements are to each other. It could
be the same device making repeated measurements under the same conditions, or
several of the same type of device making measurements at the same place and time.

What is standard deviation?

Standard deviation is a statistical
measure of precision to describe how
spread out numbers are. It is
calculated by taking the square root of
the variance.
And what is variance?
Variance is the average of the squared
differences from the mean. Mean is
just another word for average.

43
 Additional Data Quality Information
How do I calculate variance and standard deviation?
There are many online resources that explain how to calculate variance and
standard deviation. Here’s one example: Variance and Standard Deviation
Here’s a summary:
1. Calculate the average of the measurements. That’s the mean.
2. Calculate the difference between the mean from step 1 and each individual
measurement.
3. Square each of those numbers from step 2. That just means multiply each
number by itself.
4. Add all those squared values from step 3 and divide by one less than the number
of values. That means if there are “N” measurements, you divide by “N-1”. Now
you have the sample variance. (Your data set is just a sample of what could be
measured.)
5. Calculate the square root of the sample variance from step 1. Now you have the
sample standard deviation.
Software programs like Excel that can do these calculations, so
you don’t have to do all these steps by hand!
44
 Conclusion

We hope you found this presentation useful in planning for your

evaluation of low-cost sensors by collocating them with reference
monitors. More information on sensor evaluation and applications
can be found on the Air Sensor Toolbox web page. Good Luck!

This document has been reviewed by the U.S. Environmental Protection Agency, Office of Research and Development, and
approved for publication. Any mention of trade names, products, or services does not imply an endorsement by the U.S.
Government or the U.S. Environmental Protection Agency. The EPA does not endorse any commercial products, services,
or enterprises.
45