High Evaporative Emission Investigation
Field Study
United States
Environmental Protection
^1 *mAgency
-------
High Evaporative Emission Investigation
Assessment and Standards Division
Office of Transportation and Air Quality
U.S. Environmental Protection Agency
Prepared for EPA by
Eastern Research Group, Inc.
EPA Contract No. EP-C-17-011
Work Assignment No. 2-23
This technical report does not necessarily represent final EPA decisions
or positions. It is intended to present technical analysis of issues using
data that are currently available. The purpose in the release of such
reports is to facilitate the exchange of technical information and to
inform the public of technical developments.
NOTICE
4>EPA
United States
Environmental Protection
Agency
EPA-420-R-22-003
January 2022
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table of Contents
Page
1.0 Arvada: Introduction to Vapor Refueling Emissions Study 1-1
2.0 Arvada: Data Collection Methods and Procedures 2-1
2.1 Rebellion Photonics Gas Cloud Imaging Camera 2-2
2.2 Reference Vehicle Metered Releases 2-6
2.3 Gas Station Fuel Pump Transaction Data 2-11
2.4 Video Cameras 2-11
2.5 Weatherstation 2-12
2.6 Colorado Vehicle Registration Information 2-12
2.7 iPad Data Collection System 2-15
2.8 Gas Station Logsheets 2-19
2.9 Refueling Event Listing 2-20
2.10 SharePoint Database 2-22
3.0 Arvada: Analysis of Refueling Data 3-1
3.1 Characteristics of the Sampled Vehicles 3-1
3.2 Analysis of Reference Vehicle Artificial Refueling Emissions Releases 3-13
3.3 Potential Effect of Open Driver's Door on Plume Observations 3-18
3.4 Model Year Trends in Phase 1 Plume Observations 3-19
3.5 Phase 2 Re-Viewing Refueling Videos for Time Trends of Refueling Plumes 3-25
3.6 Evaluation of GCI Camera Sensitivities in Phase 2 Plume Observations 3-30
3.7 Model Year Trends in Phase 2 Plume Observations 3-41
3.8 Investigation of Refueling Plumes from Medium-Duty Vehicles 3-66
4.0 Thornton: Introduction to Liquid Refueling Emissions Study 4-1
5.0 Thornton: Data Collection Methods 5-1
6.0 Thornton: Data Processing and Database Assembly 6-1
7.0 Thornton: Results 7-1
8.0 Thornton: Analysis 8-1
9.0 Concluding Thoughts for Future Consideration 9-1
List of Appendices
Appendix A Arvada: Reference Vehicle Test Conditions and Results
Appendix B Arvada: Phase 1 Enhanced MidWave Video Viewing Instructions
Appendix C Arvada: Phase 2 Enhanced MidWave Video Viewing Instructions
Appendix D Thornton: Clicks, Spills, Spitback Data Collection Instructions
Appendix E Estimation of Headspace Vapor Properties for Denver Summer Conditions
in
-------
High Evaporative Emissions Investigation Field Study Final Report
List of Figures
Page
Figure 1-1. Arvada Gas Station Site Used for Refueling Emissions Testing 1-2
Figure 1-2. Diagram of Arvada Gas Station Islands and Pumps 1-2
Figure 2-1. Close-up of the Rebellion Photonics GCI Camera 2-3
Figure 2-2. GCI Camera Installation on Telescoping Mast on the Pick-up Truck 2-3
Figure 2-3. Rebellion GCI Camera System Viewed from Arvada Gas Pumps 2-4
Figure 2-4. Rebellion GCI Camera on Telescoping Mast 2-4
Figure 2-5. Reference Vehicle Providing Metered Artificial Refueling Releases 2-8
Figure 2-6. Windshield Pinwheel Used to Show Reference Vehicle Test Condition 2-8
Figure 2-7. Reference Vehicle Set-Up for Butane Releases 2-10
Figure 2-8. Reference Vehicle Set-Up for HeadSpace Vapor Releases 2-10
Figure 2-9. Weather Station Sensor Array 2-13
Figure 2-10. Weather Station Receiver, Display, and Logger Module 2-13
Figure 2-11. Refueling Vehicle Data Collection Interface for iPad 2-16
Figure 2-12. Example SharePoint Refueling Event Selected for Review 2-22
Figure 3-1. Model Year Distribution of Sampled Vehicles 3-4
Figure 3-2. Empty Weight Distribution of Sampled Vehicles 3-5
Figure 3-3. Fuel Tank Capacity of Sampled Vehicles 3-6
Figure 3-4. Canister Capacity Distribution of Sampled Vehicles 3-7
Figure 3-5. Canister Size vs. Fuel Tank Capacity for Confirmed ORVR Vehicles 3-9
Figure 3-6. Percent Refueling vs. Model Year for Sampled Vehicles 3-11
Figure 3-7. Distribution of Refueling Percentage for Sampled Vehicles 3-12
Figure 3-8. Trends of Plume Visibilty Probability by Release Location 3-16
Figure 3-9. Determination of Detection Limit by Release Location 3-18
Figure 3-10. Phase 1 Model Year Trend of Observable Plumes, Puffs, and Puddles 3-21
Figure 3-11. Phase 1 Model Year Trend of Front Pump (5, 6, 9) Plumes 3-23
Figure 3-12. Phase 1 Model Year Trend of Rear Pump (7, 8, 11) Plumes 3-23
Figure 3-13. A Frame with a Topping-Off Puff in a Video for the Rear Vehicle 3-27
Figure 3-14. A Frame with a Continuous Plume in a Video for the Front Vehicle 3-27
Figure 3-15. Under-Canopy Wind Speed during Confirmed ORVR Refuelings 3-33
Figure 3-16. Under-Canopy Temperature during Confirmed ORVR Refuelings 3-34
iv
-------
High Evaporative Emissions Investigation Field Study
Final Report
List of Figures (Continued)
Page
Figure 3-17. Gallons Dispensed during Confirmed ORVR Refuelings 3-35
Figure 3-18. Hour of the Day for Confirmed ORVR Refuelings 3-36
Figure 3-19. Vehicle Class for Confirmed ORVR Refuelings 3-37
Figure 3-20. Vehicle Make for Confirmed ORVR Refuelings 3-38
Figure 3-21. Pump to Fuel Fill Door Distance for ORVR Refuelings 3-39
Figure 3-22. Model Year Trend of ContinuousPlume Fraction for LDGVs 3-46
Figure 3-23. Model Year Trend of OnlyPuffsNoPlumes Fraction for LDGVs 3-46
Figure 3-24. Model Year Trend of ContinuousPlume Fraction for 0-6,000 lb GVWR trucks
(LDGT12s) 3-47
Figure 3-25. Model Year Trend of OnlyPuffsNoPlumes Fraction for 0-6,000 lb GVWR
trucks (LDGT 12s) 3-47
Figure 3-26. Model Year Trend of ContinuousPlume Fraction for 6,001-8,500 lb. GVWR
(LDGT34s) 3-48
Figure 3-27. Model Year Trend of OnlyPuffsNoPlumes Fraction for 6,001-8,500 lb.
GVWR (LDGT34s) 3-48
Figure 3-28. Overlaid Model Year Trend of ContinuousPlume Fraction 3-49
Figure 3-29. Overlaid Model Year Trend of OnlyPuffsNoPlumes Fraction 3-49
Figure 3-30. Model Year Trend of ContinuousPlume Fraction for Combined LDGVs,
LDGT 12s, and LDGT34s 3-50
Figure 3-31. Model Year Trend of OnlyPuffsNoPlumes Fraction for Combined LDGVs,
LDGT 12s, and LDGT34s 3-50
Figure 3-32. Regression of Continuous Plume Probability against Model Year and Pump
Position 3-51
Figure 3-33. Refueling Emission Profiles with ContinuousPlumes for LDGVs 3-55
Figure 3-34. Refueling Emission Profiles with ContinuousPlumes for LDGT12s 3-59
Figure 3-35. Refueling Emission Profiles with ContinuousPlumes for LDGT34s 3-62
Figure 4-1. Costco Thornton Site Used for Spills Evaluations 4-2
Figure 4-2. Diagram of Costco Thornton Gas Station Islands and Pumps 4-2
Figure 8-1. Logistic Regression Total Spill Probability vs. Extra Clicks 8-4
Figure 8-2. Logistic Regression Cumulative Spill Probability vs. Extra Clicks 8-5
Figure 8-3. Elapsed Times between Adjacent Refueling Observations 8-7
-------
High Evaporative Emissions Investigation Field Study
Final Report
List of Tables
Page
Table 2-1. Test Conditions for Butane Releases 2-9
Table 2-2. Test Conditions for Head Space Vapor Releases 2-9
Table 2-3. iPad Keystroke Event Codes Written to Microsoft SQL Server Database 2-18
Table 2-4. Sample Data in Microsoft SQL Server Database as Uploaded from the iPad
Interface 2-19
Table 3-1. Description of Gasoline Vehicle Classes 3-2
Table 3-2. ORVR Equipment in the Arvada Sample 3-3
Table 3-3. Video Plume Visibility Responses to Test Vehicle Conditions 3-13
Table 3-4. Video Plume Visibility Probabilities for Test Vehicle Conditions 3-15
Table 3-5. Estimated Effect of Open vs. Closed Door on Plume Observability 3-19
Table 3-6. Phase 1 Status of Refueling Events that Met Selection Criteria 3-20
Table 3-7. Phase 1 Status of Refueling Events for Confirmed ORVR Vehicles" 3-28
Table 3-8. Phase 2 Codes Used to Characterize Video 5-second Blocks 3-29
Table 3-9. Continuous Plumes Observed at Front and Back Row Pumps 3-31
Table 3-10. Puff Types Observed at Front and Back Row Pumps 3-31
Table 3-11. Any Puffs Observed at Front and Back Row Pumps 3-31
Table 3-12. Continuous Variable Split Values 3-40
Table 3-13. Logistic Regression Results for Wind Speed and Pump Position 3-41
Table 3-14. Phase 2 Status of Refueling Events for Confirmed ORVR Vehicles' 3-43
Table 3-15. Phase 2 Viewing Results for 2019 Light-Duty Vehicles 3-53
Table 3-16. Refueling Plume Results for HDGV2b 3-68
Table 3-17. Refueling Plume Results for HDGBs 3-70
Table 3-18. Refueling Plume Results for HDGV3s 3-70
Table 3-19. Refueling Plume Results for HDGV4s 3-71
Table 6-1. Two Transcribed Datasheet Entries 6-2
Table 7-1. Spill Occurrence Frequency for Sequential Observations 7-1
Table 7-2. Spill Size Frequency for Sequential Observations 7-1
Table 7-3. Spitback Frequency for Sequential Observations 7-1
Table 7-4. Pump Number Frequency for Sequential Observations 7-2
Table 7-5. Fueling Side Frequency for Sequential Observations 7-3
Table 7-6. Nozzle Orientation Frequency for Sequential Observations 7-3
vi
-------
High Evaporative Emissions Investigation Field Study
Final Report
List of Tables (Continued)
Page
Table 7-7. Extra Clicks Frequency for Sequential Observations 7-3
Table 7-8. Idling Frequency for Sequential Observations 7-4
Table 8-1. Fueling Side Frequency v. Total Spills and Refuelings for Sequential
Observations 8-1
Table 8-2. Nozzle Orientation Frequency v. Total Spills and Refuelings for Sequential
Observations 8-2
Table 8-3. Spill Rate v. Extra Clicks for Sequential Observations 8-2
Table 8-4. Extra-Clicks Frequency v. Total Spills and Refuelings for Sequential
Observations 8-3
Table 8-5. Extra Clicks Frequency by Spill Size for Sequential Observations 8-4
Table 8-6. Bucket-Sized Spill Frequency for Station-Wide Refuelings 8-8
Table 8-7. Spitback Frequency for Station-Wide Refuelings 8-8
vii
-------
High Evaporative Emissions Investigation Field Study
Final Report
Acknowledgments
This project demanded major commitments by several organizations and their staffs. We
needed to test at high-volume gas stations in the Denver area. We thank Tim Hurlocker, director
of Costco Wholesale's nation-wide gas station network, who got his management's approval for
Costco participation, provided access to local Costco stations, their staff, as well as coordination
with Costco's gas pump transaction information technology staff. Colorado Department of
Public Health and Environment staff, including Rob Dawson, Jim Sidebottom, Jim Kemper, and
Mike Mallory, provided local Denver assistance with gas station selection, reference test vehicle
set-up and operation, and on-site oversight during the 15-hour test days over three weeks. James
Ashby and his colleagues at PG Environmental provided critical contributions for development
of and data collection with the iPad app used at the gas station. James also created and
maintained the SharePoint database, which made click-and-view of the videos easy to do for data
quality checking and data analysis. Naima Swisz-Hall of the U.S. Environmental Protection
Agency looked up vehicle fuel tank capacities and evaporative emissions control system canister
capacities using VIN stems, model year, make, and model descriptions that ERG provided. This
project generated several large datasets of disparate data. Cindy Palacios (ERG) pulled the
datasets together, time-aligned them, and created data summaries for use by video-viewing and
data analysis staff. Finally, Amy Allen of Rebellion Photonics coordinated their gas cloud
imaging (GCI) camera activities, staff, and equipment to meet the needs of this project.
-------
High Evaporative Emissions Investigation Field Study
Final Report
Executive Summary
Light-duty vehicle onboard diagnostic (OBD) systems monitor many of the components
of emissions control systems to help ensure that those systems continue to operate as designed.
However, today's OBD systems do not monitor the performance of the evaporative emissions
control system canister, which captures and stores evaporative emissions for subsequent
combustion in the engine. The U.S. Environmental Protection Agency (EPA) has been
conducting a series of studies to determine if this approach is justified.
With the 1998 model year, regulations began to be phased in that required manufacturers
to equip vehicles with evaporative emissions control systems that would reduce evaporative
emissions generated during vehicle refueling. Refueling emissions are generated when the liquid
gasoline put into the fuel tank displaces the tank's headspace vapor. The composition of
headspace vapor varies with gasoline volatility and fuel tank temperature, but as a rule of thumb,
headspace vapor is approximately 50 vol% hydrocarbon. Pre-1998 gasoline vehicles simply vent
this vapor to the atmosphere. Typically, to help control refueling emissions, the new onboard
refueling vapor recovery (ORVR) evaporative emissions control systems create a liquid seal by
extending the fuel fill pipe inside the fuel tank to near its bottom. By various methodologies
using different designs, the gasoline vapor is routed to the canister for capture and is prevented
from returning through the fill pipe.
While today's ORVR systems are required to also control diurnal, hot-soak, and running
loss evaporative emissions, they are most severely challenged by refueling emissions because a
large mass of headspace vapor from the vehicle gas tank must be controlled over the short period
when refueling occurs. This study takes advantage of this fact, and of the fact that all vehicles
must refuel, by monitoring the refueling vapor emissions of a sample of the fleet at a commercial
gas station.
Another source of refueling emissions is liquid gasoline spills and leaks from vehicles.
These liquid sources are related to the behaviors of gasoline station customers, vehicle
maintenance, and gas pump nozzle design. Accordingly, this study also collected data on the
characteristics of liquid gasoline spills and leaks at a commercial gas station.
Overall, this study collected and analyzed gasoline station refueling data to evaluate
gaseous and liquid refueling emissions on light- and medium-duty gasoline-fueled vehicles.
Costco Wholesale participated in the study by allowing us to collect data at two of their gas
stations in the Denver area from July 7 to 23, 2019. These gas stations pumped only gasoline,
which had 9.0 psi RVP volatility and 10 vol% ethanol during the study period. The gaseous
IX
-------
High Evaporative Emissions Investigation Field Study
Final Report
refueling emissions data was collected at the Arvada Costco gas station, and liquid refueling
observational data was collected at the Thornton Costco gas station.
Data Collection of Refueling Vapor Emissions - At the Arvada station, we used the
Rebellion Photonics gas cloud imaging (GCI) infrared hyperspectral video camera and video
post-processing to identify refueling emission puffs and plumes on 2,854 vehicle refuelings. The
gas station had two pumps on each side of three islands. On each day, the infrared camera was
positioned about 110 feet in front of and about 30 feet above the two refueling positions on one
side of an island to allow videoing of two vehicles refueling. Simultaneously, we recorded
vehicle arrival and departure times and used license plates and a snapshot of the Colorado
registration database to confirm vehicle identity while the vehicle was still there. Costco
provided timestamps of pump nozzle lift-off and hang-up times and volume of fuel dispensed.
EPA looked up fuel tank capacity and canister capacity for most of the Colorado vehicles
videoed.
At the gas station, Colorado Department of Public Health and Environment (CDPHE)
emission technical laboratory personnel produced 108 10-gallon/minute metered releases of
known concentrations of butane and gasoline headspace vapor from three different locations on a
CDPHE reference vehicle to determine the sensitivity of the infrared camera under different
viewing conditions.
The various datasets from the Arvada station measurements were merged and time
aligned to the nearest second. We viewed the 8,462 30-second infrared videos to identify
refueling emissions occurrences from private vehicles and CDPHE test runs. Each private
vehicle refueling was rated in an initial effort as either 0 (no emissions visible), L (low-density
emissions visible), H (high-density emissions visible), or P (emissions from a puddle on the
pavement).
Infrared Camera Detection Limits - Properly operating in-use ORVR vehicles should
have refueling emissions below the standard of 0.2 gHC/gallon. At the other extreme, non-
ORVR vehicles or ORVR vehicles with inoperative evaporative emissions control systems
should have refueling emissions that have concentrations equal to the fuel tank headspace HC
concentration. For the environmental and fuel properties at the time of this study, we estimate the
headspace concentration would be about 4.6 gHC/gallon. Vehicles with partially operating
control systems will have refueling emissions rates between 0.2 and 4.6 gHC/gallon. Of course,
vehicles with fuel system liquid leaks, which we did see in the study, could have refueling
emissions that are greater than 4.6 gHC/gallon.
x
-------
High Evaporative Emissions Investigation Field Study
Final Report
Analysis of the CDPHE reference vehicle releases indicated that the probability of
observing a refueling emission in a video depended on the location of the release on the vehicle
and the hydrocarbon concentration in the release. The analysis indicated that, regardless of the
vehicle release location, refueling emissions of properly operating ORVR vehicles (less than 0.2
gHC/gallon) probably would not be seen in GCI camera videos, but refueling emissions of non-
ORVR or inoperative ORVR systems (about 4.6 gHC/gallon) would very likely be seen in GCI
camera videos. This is good emissions detection behavior for the study since, in general, the
videos will be able to distinguish between control systems with good and poor behavior. The
probability of seeing refueling emissions in GCI videos of vehicles with partial control will vary
depending on emissions release location, actual emissions rate, scene composition and
illumination conditions.
Refueling Emissions Detection for Pre-ORVR Vehicles - We used the VINs looked up
in the Colorado vehicle registration database to determine the vehicle classes and ORVR
equipment of the refueling vehicles. Our analysis of the Colorado-registered vehicles indicated
that about 90% of the vehicles refueling at the Costco Arvada gas station during this study were
equipped with ORVR systems. Vehicles with model years before the start of the phase-in to full
ORVR implementation should all produce observable refueling emissions. Preliminary analysis
(see Table 3-6) of the 2,854 refuelings videoed showed that 82 refuelings were for pre-1998
(unambiguously pre-ORVR) vehicles, and 70 (85%) of those showed refueling emissions in their
videos. Based on the GCI camera detection limits, which were discussed above, we would expect
100%) of pre-ORVR vehicles to exhibit plumes in the videos. However, several factors can
contribute to reducing the probability of observing emissions in videos of refueling events. These
factors include the amount of fuel dispensed, the calibration and on/off status of the GCI camera
at the time of the refueling, obstacles (other vehicles, people, car doors) in the line of sight,
background illumination, and wind speed. No further analysis of pre-ORVR vehicle refueling
emissions was undertaken.
Refueling Emissions for Confirmed ORVR Vehicles - The analysis of the Arvada
station data concentrates on 1,990 refueling events of vehicles built after the respective model
year of the full ORVR implementation, that is, 2000 MY for light-duty gasoline vehicles
(LDGV), 2003 for light-duty gasoline trucks with gross vehicle weight less than 6,000 pounds
(LDGT12), and 2006 for light-duty trucks with GVWR between 6,000 and 8,500 pounds
(LDGT34). The above-mentioned initial video examinations revealed that emissions were rarely
observed during the entire refueling event, but that emissions started and stopped. Additionally,
emissions were made up of puffs and plumes, which we define for this study:
XI
-------
High Evaporative Emissions Investigation Field Study
Final Report
Puffs are short-term emission events associated with the removal of the gas cap, the
beginning of fuel flow into the fuel fill pipe, the end of fuel flow when the pump nozzle
automatically clicks off, customer's efforts to top off the fuel tank using extra nozzle
clicks, or emissions from puddles on the ground produced by vehicle fuel line leaks or
drips from the nozzle when it is transferred between the pump and the vehicle.
Plumes are generally longer-term emissions events when the fuel is flowing at a steady
rate into the fuel fill pipe.
Because of the time-varying nature of the refueling emissions and because ORVR
systems should control plumes but not necessarily puffs, we re-viewed the videos in a Phase 2
effort - but only those for confirmed ORVR vehicles that had initial video viewing results of
Low density, High density, or Puddle. In the Phase 2 video viewings, we characterized each 5-
second block of each video by giving separate characterization codes to puffs and to plumes so
that separate analyses could be done on the two emission types. The Phase 2 puff and plume
codes were used to assign each ORVR refueling event as one of three categories:
NoPuffsNoPlumes: We saw neither puffs nor plumes in any of the videos for the
refueling event. 81% (=1109/1990) of the refuelings of confirmed ORVR-equipped
vehicles were in this category.
OnlyPuffsNoPlumes: We saw at least one puff of any type (remove gas cap, begin fuel
flow, nozzle click-off, topping off, puddles), but we did not see any plumes associated
with periods of steady fuel flow. 15% (=292/1990) of the refuelings of confirmed ORVR-
equipped vehicles were in this category.
ContinuousPlumes: We saw plumes associated with periods of steady fuel flow, and
puffs of any type may or may not have been present. 3.9% (=77/1990) of the refuelings of
confirmed ORVR-equipped vehicles were in this category.
It is important to recognize that the 81%, 15%, and 3.9% percentages of refuelings that
were assigned to the three puff/plume categories depend strongly on the sensitivity of the video
camera. All refuelings produce some level of emissions. Thus, if we had used a more sensitive
video camera, plumes would have been seen for every refueling. Alternately, if we had used a
low sensitivity video camera, no emissions would have been seen at all. These considerations
should be taken into account when using this dataset for modeling.
The data collected at Arvada shows that camera sensitivity is influenced by viewing
conditions. Since the identity of the gas station pump should be independent of the ORVR
system state of repair, the fraction of ORVR vehicles that produce observable plumes at any of
the pumps should be the same within statistical uncertainty. When we tested this assumption, we
found that refuelings at the back row of pumps, which are farther from the video camera, had a
-------
High Evaporative Emissions Investigation Field Study
Final Report
60% larger fraction of observable plumes than those at the front pumps. We believe that this
difference was caused by the scene's background complexity and illumination. Back-pump
backgrounds generally had smooth pavement and were illuminated by the sun, which caused
stronger ambient infrared radiation. Front-pump backgrounds generally contained another
vehicle and moving customers and were in the shadow of the gas station canopy. We also found
that the observable plume sensitivity was enhanced by calm wind conditions.
Emissions video data that was collected under less sensitive video conditions should not
be discarded. Also, emissions video data collected under more sensitive video conditions should
not be believed more than data collected under less sensitive video conditions. Instead, all data
should be analyzed by considering the differences of camera sensitivity and by recognizing
which results trends are subject to the influence of camera sensitivity and which are not.
Major Findings of the Arvada Gas Station Study - We judged the refueling emissions
behavior of ORVR vehicles as a function of vehicle class and model year. Because continuous
plumes, but not puffs, should be controlled by ORVR systems, we judged emissions performance
based on controlling continuous plumes. Judging criteria were the model-year dependence of
continuous plume prevalence, which is the fraction of vehicles in the sample that had observable
continuous plumes.
Here are the major findings of the Arvada gas station study for prevalence of continuous
plumes for ORVR vehicles:
1) We found no statistical difference between the model year trends of continuous plumes
for LDGV, LDGT12, and LDGT34 vehicles with ORVR systems.
2) The prevalence of continuous plumes (i.e. not puffs) is near zero for new vehicles
(2019 model year in this study) that have ORVR systems.
3) As ORVR vehicles age, the occurrence of continuous plumes, as measured by
prevalence, increases approximately proportionally with vehicle age, but the mass rate of
emissions degradation is unknown.
The rate of increase of continuous plumes in older vehicles as measured in this study
depends on camera sensitivity. For example, for back pumps, where the videoing sensitivity was
higher, the prevalence increases at about 0.56%/model year. For front pumps, prevalence
increases at about 0.41%/model year. Using only the results of this study, neither of those rates
can be used directly for modeling purposes because those rates are not independent of camera
and videoing conditions. For modeling, the degradation needs to be quantified on a mass basis so
that it is independent of the measurement sensitivity.
-------
High Evaporative Emissions Investigation Field Study
Final Report
Data Collection of Refueling Liquid Emissions - At the Thornton station, we randomly
and sequentially selected vehicles arriving at the station to monitor the prevalence of liquid fuel
drips, spills, leaks, and spitbacks and the behaviors and refueling conditions associated with
them. For each of the 1,227 monitored vehicles, we recorded license plate, vehicle make and
model, gas pump number, fuel nozzle orientation, the number of extra fuel nozzle clicks that the
customer used to top off the fuel tank, the fuel nozzle hang-up time, the relative size of spills and
spitbacks, and the gas station attendant's response to them. We also recorded situational
information on extreme spill or spitback occurrences when they were noticed station-wide even
if those vehicles had not been randomly selected. Costco provided timestamps of pump nozzle
lift-off and hang-up times and volume of fuel dispensed.
Major Findings of the Thornton Gas Station Study - According to the 1,171
sequential refueling observations, 10.3% of gas station customers spilled gasoline, although most
spills were small, nickel-sized spills. About two-thirds of customers accepted the automatic shut-
off of the gas pump nozzle and did not try to top-off their tank. Only 8.4% of those refuelings
resulted in a spill. When customers attempted to top-off their gas tank by adding extra clicks to
their refueling nozzle, they tended to spill at a greater frequency. For example, 5% of all
customers used two extra clicks of the nozzle, which resulted in a spill rate of 20.6% - two and
one-half times the spill rate experienced by customers who accepted automatic nozzle shut-off
Extra clicks were also associated with larger spills. For example, 31 customers used between 10
and 18 extra clicks, and 2 of those refuelings produced large spills, which we called bucket-sized
spills to reflect the fuel volume of a small bucket (diameter of bucket and higher). In contrast,
785 refuelings had no extra clicks, which produced only 1 bucket-sized spill. Thus, using 10 to
18 extra clicks was 50 times more likely to cause a bucket-sized spill than accepting automatic
nozzle shut-off The combined sequential observations and station-wide observations establish
that the likelihoods of bucket-sized spills and spitbacks are approximately 0.4% and 0.2%,
respectively.
Overview
Because all vehicles must refuel, the EPA saw that monitoring the refueling emissions of
ORVR vehicles at gas stations was a means of conveniently evaluating the evaporative emissions
control systems of large numbers of in-use vehicles. Accordingly, EPA conducted a preliminary
two-day gas station study1 in Austin, Texas in December 2015 to evaluate the capabilities of the
1 T.H. DeFries, "Evaluation of Rebellion Photonics Gas Cloud Imaging Camera for Screening Refueling
Evaporative Emissions from Light-Duty Vehicles," prepared for U.S. Environmental Protection Agency,
prepared by Eastern Research Group, EPA-160411, April 11, 2016.
xiv
-------
High Evaporative Emissions Investigation Field Study
Final Report
Rebellion Photonics camera and a gas station pilot study at a Shell gas station in Wheat Ridge,
Colorado in November 20182.
This report documents two essentially simultaneous studies at two Costco Wholesale gas
stations - one in Arvada and the other in Thornton, Colorado. We describe the data collection,
results, and analysis of refueling emissions. To prepare for this project under Work Assignment
2-23 for Contract EP-C-17-011, Eastern Research Group (ERG) wrote a work plan3 and a quality
assurance project plan4. ERG ran the studies for EPA.
The study at Arvada, Colorado (Sections 1, 2, and 3) focuses on gasoline vapor
emissions, and the study at Thornton, Colorado (Sections 4, 5, 6, 7, and 8) focuses on gasoline
liquid emissions, which, of course, soon evaporate. The field data collection for the Arvada study
was conducted July 7-23, 2019 except for July 14 and 19, when video data was being
downloaded from the field server.
CDPHE has a cooperative research agreement with EPA. CDPHE provided resources for
the Arvada study including obtaining the Colorado registration database snapshots and
providing, equipping, and operating the reference vehicle and its artificial refueling emissions
releases. PG Environmental, which is a subsidiary of ERG, is in nearby Golden, Colorado and
was a subcontractor to ERG. For the Arvada study, they developed the iPad data collection
system, provided staff to collect data at the gas station using the iPad, and viewed the thousands
of infrared videos to search for refueling emissions plumes and other refueling features. For the
Thornton study, PG Environmental staff collected information on paper logsheets as they
observed customers refueling. For the Arvada study, Rebellion Photonics, an ERG subcontractor,
collected continuous infrared measurements using its Gas Cloud Imaging video camera and
processed the data on site to produce Enhanced MidWave videos that could reveal refueling HC
plumes. Jim Sidebottom and Jim Kemper were consultants on the study. Since they live in the
Denver area and were formerly full-time CDPHE employees, they assisted making arrangements
with CDPHE and local agencies and businesses before, during, and after the field data collection.
2 T.H. DeFries, "High Evaporative Emissions Investigation Field Study: November 2018 Pilot Study,"
draft report, prepared for U.S. Environmental Protection Agency, prepared by Eastern Research Group,
Austin, TX, EPA-190219, February 19, 2019.
3 "High Evaporative Emissions Investigation Field Study, Work Plan, Version 3," prepared for U.S.
Environmental Protection Agency, prepared by Eastern Research Group, EPA-190424, April 24, 2019.
4 T.H. DeFries. "High Evaporative Emissions Investigation Field Study, Quality Assurance Project Plan,
Version 2," prepared for U.S. Environmental Protection Agency, prepared by Eastern Research Group,
EPA-190920, September 20, 2019.
xv
-------
High Evaporative Emissions Investigation Field Study
Final Report
Costco Wholesale was a key partner in both studies. ERG approached Costco via their
corporate offices in Issaquah, Washington. At no cost to the project, Costco allowed us to work
at and collect data at their Arvada gas station and Thornton gas station. Additionally, Costco
provided the detailed timestamp and gallons dispensed data on each gas pump transaction at the
two stations during the entire data collection period. The only requirement that Costco had was
for ERG to keep the private information of its members confidential. Accordingly, ERG agreed
not to reveal Costco member vehicle license plates or VINs to anyone outside of ERG and its
subsidiary PG Environmental. Even EPA and CDPHE were not allowed to have member
confidential information. Therefore, in this document we show no license plates and show only
VIN stems.
A key EPA study requirement was to obtain all refueling event information without study
personnel initiating interactions with vehicle owners in any way or touching their vehicles. Thus,
direct access to vehicles, canisters, and OBD data was not allowed in these studies.
xvi
-------
High Evaporative Emissions Investigation Field Study
Final Report
1.0 Arvada: Introduction to Vapor Refueling Emissions Study
This study estimates the prevalence of vehicles with elevated refueling emissions as
vehicles are refueled at gas stations. The focus is on Tier 2 and Tier 3 vehicles, but the study
collected data on all technologies. The measurement equipment provides an estimate of the
distribution of relative refueling HC emissions at ground level as vehicles refuel. Elevated
refueling emissions can occur because of canister degradation, which may not be OBD-
detectable; OBD-detectable problems, which includes purge system issues and other evaporative
emissions control system problems; or simple overwhelming of a problem-free evaporative
emissions control system because of extreme environmental conditions such as elevated ambient
temperature when the vehicle contains volatile fuel or driving patterns that prevent evaporative
canister purge events. Thus, the distribution of refueling emissions are likely an upper estimate
of the emissions based on canister degradation alone.
Testing for refueling plumes was conducted at the Costco Wholesale Arvada gas station,
5195 Wadsworth Boulevard, Arvada, Colorado 80022. This gas station was open for business
Monday through Friday from 6 a.m. to 9 p.m. and on Saturday and Sunday from 7 a.m. to 7 p.m.
The gas station was available for testing during all hours that the station was open for business.
The customers at the Costco gas station must be Costco members. The methods of payments that
are accepted at the gas station are credit card and debit card only. Cash is not accepted for
gasoline purchases. The gas station pumps only gasoline and only regular and premium grades.
No diesel fuel was available at the station. Two samples of gasoline were collected for analysis
by CDPHE. The results indicated that the gasoline samples had a volatility of 9 psi RVP and
10% ethanol.
Figure 1-1 shows a Google Maps view of the Costco Arvada site with north being up in
the photograph. The figure shows the canopy, which is approximately 32 by 86 feet. Traffic
flows through the gas station from the southwest to the northeast, as shown by the arrows on the
pavement. Customers cue up between the dashed lines on the pavement just southwest of the
canopy. Traffic is not allowed to flow in the opposite direction.
1-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 1-1. Arvada Gas Station Site Used for Refueling Emissions Testing
1-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 1-2 shows the three fueling islands and the numbering of the twelve fuel pumps
under the canopy using the same orientation as Figure 1-1. The islands are far enough apart from
each other that customers can enter or leave a fuel pumping area even if another vehicle is
fueling in front of them. On any given testing day, video data was collected on either Pumps 5
and 7, Pumps 6 and 8, or Pumps 9 and 11. At the same time, a study technician collected
refueling event and vehicle information on all four pump positions at the fueling island being
monitored - either Island 5/6/7/8 or Island 9/10/11/12. The technician was safely positioned at
the center of the targeted island to get a good view of vehicles and their license plates.
Two GoPro video cameras were placed on the northeast corner of the roof of the Costco
gas station manager's building to record the movement of refueling vehicles. The location is
shown by the green square in the right center of Figure 1-1. The study's weather station was
placed on the top of the gas station manager's booth, which was in the center of the middle
island.
The truck equipped with the gas cloud imaging camera was located about 40 meters to
the northeast of Island 5/6/7/8 or Island 9/10/11/12 such that the line of sight of the camera was
approximately on the centerline of the fueling island. The two locations used by the truck are
shown in Figure 1-1 by the two yellow rectangles in the upper right-hand corner. The black
circle inside the yellow rectangles represents the location of camera and its mast.
1-3
-------
High Evaporative Emissions Investigation Field Study
Final Report
2.0 Arvada: Data Collection Methods and Procedures
The goal of the refueling emissions plume study was to collect data that could be used to
estimate the prevalence of refueling plumes and, if possible, to quantify the level of refueling
emissions. The focus was to be on Tier 2 and newer technologies, but the behavior of older
technologies was of interest and is used as a positive control for plume presence. Specifically, all
pre-ORVR vehicle refuelings are expected to produce refueling plumes. A part of the overall
goal was to estimate how the evaporative emissions control systems of ORVR vehicles degraded
with age. To meet these goals, ERG designed a test program with several components to get the
needed information and devised a way to connect the information. This section describes the
major methods that we used.
Rebellion Photonics Gas Cloud Imaging Camera (Section 2.1) - This 15-frame-per-
second video camera measures infrared radiation and separates it into wavelength bands.
Rebellion's special processing techniques can convert the data into enhanced videos
where refueling plumes can be seen.
Reference Vehicle Metered Releases (Section 2.2) - CDPHE provided and set up a
reference vehicle that released various concentrations of simulated refueling HC
emissions from different release points. These releases, which were done at the same gas
station pumps where private vehicles refueled, were used as a reference forjudging the
refueling emissions of the private vehicles in the fleet.
Gas Station Fuel Pump Transaction Data (Section 2.3) - Costco provided transaction
data (scrubbed for private information) for every gasoline purchase during the study
period. This included timestamps for credit card approval and nozzle hang-up and for
volume of fuel dispensed.
Video Cameras (Section 2.4) - We photographed the gas station scene with video and
high-resolution time-lapse cameras to document the movement of vehicles during data
collection.
Weather Station (Section 2.5) - We installed a weather station to record temperature
and the speed and direction of wind under the gas station canopy, which is where these
environmental factors affect dispersion of refueling emissions plumes.
Colorado Vehicle Registration Information (Section 2.6) - CDPHE provided us with
vehicle descriptions of all vehicles registered in Colorado so that we could confirm the
identity of refueling vehicles via their license plates.
iPad Data Collection System (Section 2.7) - We developed and used, at the gas station,
a custom iPad app that was linked to the cloud via a mobile Wi-Fi hotspot to collect and
store timestamps for vehicle arrival and departure and to look up vehicle descriptions via
observed license plates to visually confirm vehicle identity.
2-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
Gas Station Logsheets (Section 2.8) - We used paper logsheets to supplement the iPad
data collection system with information on less common vehicle refueling behavior, such
as refueling gas cans, lawn mowers, and jet skis; refueling multiple vehicles during a
single gas pump transaction; and vehicle descriptions that were not found in the Colorado
vehicle registration snapshot.
Refueling Event Listing (Section 2.9) - We wrote a SAS program that time-merged all
of the previous data sources to provide a second-by-second chronological record of all
refueling information for the four pumps at the gas station island where data was
collected each day. Then, the program converted the chronological record into a listing
where each event's refueling information was included in a single observation.
SharePoint Database (Section 2.10) - The refueling event listing was pulled into a
Microsoft SharePoint database stored in the cloud. The database was a convenient tool
for quality-checking, sorting, and viewing - including playing Rebellion videos - of all
results of data collection for the gas station study.
2.1 Rebellion Photonics Gas Cloud Imaging Camera
For this project, Rebellion Photonics used its Gas Cloud Imaging (GCI) camera to screen
for gasoline vapor being emitted from refueling vehicles in their surroundings. This camera is an
infrared hyperspectral camera that collects video data. We had used the camera in a previous
study5 in Austin, Texas, and in the refueling emissions pilot study, which was conducted at a
Shell gas station in Wheat Ridge, Colorado, in November 20186.
The GCI camera is part of a system constructed and operated by Rebellion Photonics.
This system is made up of a pickup truck equipped with computers and associated electronics to
store data and a vertical telescoping mast supporting the camera. In this study the mast was
elevated to a height of about 30 feet. Figure 2-1 shows a close-up of the camera, and Figure 2-2
shows the camera system including the pick-up truck. Figure 2-3 shows a photograph of the
camera system as viewed from one of the refueling position gas pumps in the study. Figure 2-4
shows a close-up of the mast with the camera on top viewed from the side of the pickup truck.
The junction between the top of the mast and the bottom of the camera is fitted with a mount that
can pan and tilt to change the vertical and horizontal view of the camera. The camera does not
5 T.H. DeFries, "Evaluation of Rebellion Photonics Gas Cloud Imaging Camera for Screening Refueling
Evaporative Emissions from Light-Duty Vehicles," prepared for U.S. Environmental Protection Agency,
prepared by Eastern Research Group, Austin, Texas, EPA-160411, April 11, 2016.
6 T.H. DeFries, "High Evaporative Emissions Investigation Field Study: November 2018 Pilot Study,"
draft report, prepared for U.S. Environmental Protection Agency, prepared by Eastern Research Group,
Austin, TX, EPA-190219, February 19, 2019.
2-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 2-2. GCI Camera Installation on Telescoping Mast on the Pick-up Truck
have the capability of zooming, and therefore the size of images can be adjusted only by
changing the distance between the camera and the object that is being viewed.
Figure 2-1. Close-up of the Rebellion Photonics GCI Camera
2-3
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 2-3. Rebellion GCI Camera System Viewed from Arvada Gas Pumps
Figure 2-4. Rebellion GCI Camera on Telescoping Mast
-------
High Evaporative Emissions Investigation Field Study
Final Report
The GCI camera measures the intensity of infrared radiation in the scene across the pixels
of the camera sensor array. The infrared radiation falling on each pixel is separated into
measurements in 10 to 15 wavelength bands between 3 and 14 micrometers. The video frame
rate of the camera is 15 frames per second. The GCI camera generates raw data at approximately
1GB per minute. This data is stored on large hard drives located inside the cab of the pickup
truck.
For this study, the GCI camera collected raw data continuously during almost all hours of
gas station operation. The camera was aimed at one side of a chosen island for the day such that
the front few inches of the gas pumps were on one side of the frame and the other side of the
frame was on the far side of the vehicle being refueled. The GCI camera was elevated as high as
possible so that video could be simultaneously obtained on the vehicle at the near pump and the
vehicle at the far pump. The raw data was processed on site in 30-second blocks. One 30-second
block of raw data was processed to produce a 30-second black and white Enhanced MidWave
video. This video was stored on the hard disk. Thus, the hard disk contained all of the continuous
raw data and a series of 30-second Enhanced MidWave videos. Because the processing of the
30-second block of raw data takes approximately 10 seconds, the 30-second Enhanced MidWave
videos are separated by at least 10 seconds to allow time for processing.
While the raw data contains infrared measurements from 3.2 to 3.5 and 7.5 to 14 |im, the
Enhanced MidWave videos are created only from the measurements obtained in the 3.2 to 3.5
|im band, which is the infrared region where most hydrocarbon molecules absorb infrared
radiation.
The GCI camera produces infrared videos of hydrocarbon vapor emissions by imaging
the ambient infrared radiation in the scene. Where hydrocarbon vapor is not present, the infrared
measurements show the radiation from the background. However, when hydrocarbon vapor is
present in the scene, the vapor absorbs some of the infrared radiation that is being emitted by the
background, and the absorption is a function of time and space across the scene. This produces
time-varying contrasts or discontinuities in the infrared video images. Because the camera
produces video, the movement of these discontinuities is perceived by a person viewing the
video as a cloud that is moving in the scene.
The GCI camera is recalibrated every 8 minutes by inserting a white card in the optical
path. Recalibration is necessary because the sensor array can become saturated by strong infrared
radiation, for example, from the reflection of sunlight from various shiny objects in the scene.
2-5
-------
High Evaporative Emissions Investigation Field Study
Final Report
Calibration is performed automatically inside the GCI camera. It results in a white screen in the
Enhanced MidWave videos.
At the end of each day of testing at the Arvada station, Rebellion Photonics personnel
downloaded the Enhanced MidWave video files onto a thumb drive for archiving by ERG
personnel. Rebellion personnel did not provide the raw GCI data. Two days were needed during
the two weeks to download the data from the field servers, therefore no data was collected on
July 14 or 19 during the study period July 7-23.
2.2 Reference Vehicle Metered Releases
During several days of GCI camera data collection, CDPHE provided a reference vehicle
and released metered amounts of known hydrocarbon vapor concentrations. The objective was to
use the metered releases as controls to judge the concentration of hydrocarbons emitted by
refueling vehicles.
Two types of HC gases were released as artificial refueling emissions: butane, and
gasoline headspace vapor. Mixtures of these gases with nitrogen were used to simulate refueling
emissions. The artificial refueling releases were released from a reference vehicle that had low
evaporative emissions of its own. Releases were made at the participating gas station at the same
gas-pump locations used by private vehicles whose emissions were being monitored by the GCI
camera.
The total flow of the artificial releases was 10 gallons/minute, which is a typical fuel
dispensing flow of gas station fuel pumps. According to our ReddyEvap 2010 calculations,
headspace vapor in Denver at summer temperatures is approximately 50 vol% HC vapor7. We
used 10 gallons/minute and 50 vol% HC as the basis for determining artificial refueling
emissions release flow and composition. Reference vehicle hydrocarbon vapor release mixtures
were produced to simulate 10%, 30%, and 100% of the equilibrium gasoline headspace
concentrations (i.e., of 50 vol% HC).
Figure 2-5 shows a photograph of the reference vehicle, a 2003 Ram pickup truck.
CDPHE fitted the pickup truck with a butane tank, a gasoline caddy, and a pressurized cylinder
of nitrogen to provide hydrocarbon / nitrogen mixtures to one of three vehicle locations: the
7 We made ReddyEvap 2010 headspace calculations using the following inputs: 8.7 psi RVP fuel, 10
vol% ethanol in the fuel, 88 F ambient temperature, 0.83 atm barometric pressure. The partial pressures
were: ethanol 62.53 mmHg, non-ethanol HC 289.96 mmHg. The barometric pressure was 631 mmHg
(=0.83 * 760 mmHg). Therefore, the headspace composition was: ethanol 10 vol%, non-ethanol HC 46
vol%, and air 44 vol%.
2-6
-------
High Evaporative Emissions Investigation Field Study
Final Report
actual fuel fill door on the left side of the vehicle, an artificial fuel fill location on the right side
of the vehicle behind the passenger door, and the top of the fuel tank underneath the vehicle.
We needed to have a foolproof way to designate the reference vehicle test condition by
including some physical element in the video scene. Test conditions were defined by gas
released (butane, or gasoline headspace vapor), relative HC release concentration (100%, 30%,
or 10%)), and release point (left fuel fill door, right fuel fill door, or top of tank). The key
challenge is that conventional printed text cannot be read in the infrared; we needed a shape that
could be clearly seen in the infrared videos. Our solution was a pinwheel, whose orientation
when placed on the test vehicle windshield, as shown in Figure 2-6 indicated the test condition.
One side of the pinwheel was for butane releases and the other side for gasoline headspace vapor
releases. The pinwheel had six rotational orientations that indicated the six combinations:
100/Door, 030/Door, 010/Door, 100/Tank, 030/Tank, and 010/Tank. To be able to distinguish
the orientation, we placed a plastic container filled with ice, which has low infrared emissions, at
a specified location on the pinwheel. Finally, the pinwheel was placed under the windshield
wiper on the passenger's side, in the center, or on the driver's side to designate releases from the
passenger-side fuel fill door, the top of the gas tank, or from the driver-side fuel door,
respectively. For example, even though the text cannot be read on the pinwheel in Figure 2-6, the
pinwheel location and configuration indicate that the test condition is for Butane/030/Tank.
Appendix A shows a listing of the successfully created reference vehicle test conditions
and results.
2-7
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 2-5. Reference Vehicle Providing Metered Artificial Refueling Releases
2-8
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 2-7 shows a piping diagram for the butane / nitrogen mixing and flow regulation.
Three rotameters on the outlet of the butane tank control the butane flow, a single rotameter on
the outlet of the nitrogen cylinder control nitrogen flow, and a diverter valve is provided to send
the mixture to either the left fuel fill door, the right fuel fill door, or the top of the gas tank. The
flows of butane / nitrogen were adjusted so that the total flow of the mixture was maintained at
10 gallons per minute, which is the approximate refueling flow of the gas pumps at the station.
Table 2-1 shows the flows of the components used to produce the butane mixtures.
Table 2-1. Test Conditions for Butane Releases
Relative
Carrier
Total
Test
Headspace
Butane
(i.e. N2)
Gaseous
Condition
HC Mass
Flow
Flow
Release
Name
(%)
(gal/min)
(gal/min)
(gal/min)
BUT 100
10U
5.0
5.0
10.0
BUT030
30
1.5
5.0
6.5
BUT010
10
0.5
5.0
5.5
Figure 2-8 shows the reference vehicle setup for releases of headspace vapor. Gasoline
from the plastic caddy, which was placed in the bed of the pickup truck, was pumped through a
flow meter to the vehicle's gas tank. The inlet at the fuel fill pipe was sealed so that vapor from
the vehicle's fuel tank was not allowed to come out the fuel fill pipe. Instead, the vapor from the
top of the vehicle fuel tank was routed to the tee at the exit of the nitrogen cylinder. From there
the mixture was routed to either of the fuel fill doors or the top of the gas tank underneath the
vehicle. This arrangement allowed mixtures of 10% and 30% gasoline headspace vapor in
nitrogen. To produce 100% fuel headspace vapor concentrations, gasoline was pumped directly
from the Costco fuel pump into the test vehicle fuel tank. Table 2-2 shows the flows of the
components used to produce the headspace vapor mixtures.
Table 2-2. Test Conditions for Head Space Vapor Releases
Relative
Gasoline
Carrier
Total
Test
Headspace
Flow into
Liquid
(i.e. N2)
Gaseous
Condition
HC Mass
Fuel Tank
Gasoline
Flow
Release
Name
(%)
(gal/min)
Source
(gal/min)
(gal/min)
GAS 100
100
10.0
Slnlion »;is pump
0.0
lo.o
GAS030
30
3.0
Caddy
3.5
6.5
GAS010
10
1.0
Caddy
4.5
5.5
During releases of hydrocarbon / nitrogen mixtures from the reference vehicle, Rebellion
personnel provided feedback to ensure that the raw data had been successfully collected by the
GCI camera.
2-9
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 2-7. Reference Vehicle Set-Up for Butane Releases
FF Door
• Tank Top
Figure 2-8. Reference Vehicle Set-Up for HeadSpace Vapor Releases
2-10
-------
High Evaporative Emissions Investigation Field Study
Final Report
2.3 Gas Station Fuel Pump Transaction Data
Refueling events of large volumes of dispensed gasoline are potentially more likely to
have larger refueling emissions than smaller volumes. Additionally, timestamps of transactions
at each pump help confirm the timing of refueling activities. Costco provided pump transaction
data with pump number, dispensed fuel grade and volume, date and time of credit card validation
which occurs before refueling can begin, and the date and time of fuel nozzle hang-up for all
transactions at the Arvada gas station during the data collection period.
2.4 Video Cameras
The entire gas station refueling area was filmed from a single fixed perspective during the
hours that data collection took place. Two cameras were used, one for taking continuous video of
the station, and one that took a still image every 10 seconds. The cameras used were GoPro Hero
5 cameras, mounted to a custom-built plate with gimbaled camera attachment points used to
easily adjust the fields of view for each camera. The cameras were placed on top of the
attendant's monitoring building near the edge of the fueling station area.
To allow for the cameras to run continuously for 8-12 hours at a time, rechargeable USB
batteries were used so that cameras could be plugged in to give them long run times. Batteries
were charged overnight, and extra batteries were always in reserve for backup or if a battery
could not be charged in time overnight.
The GoPro cameras could be controlled wirelessly and remotely via an iPad application
that allowed for a live view of what the camera saw, adjustment of settings, and the starting and
stopping of filming. This application was used to confirm the correct field of view for the
cameras when positioning them on top of the building and to periodically check the units and
ensure that they were still running and not out of storage space.
The GoPro Hero 5 model is capable of 12 mega-pixel still images and up to 4K resolution
live video. For this project, the still images captured every 10 seconds were taken at 4000 x 3000
pixels and 72 dpi. Live video was filmed using 1920 by 1440 pixels to maximize storage space
and reduce the heat created by the camera when continuously run.
Heat was a significant issue during filming and caused the cameras to shut down when
they reached a critical temperature. Ambient temperatures during the sampling in July were in
the 80s to 90s. A custom-built shade was constructed to shelter the cameras from direct sunlight
and reduce the heat, however the cameras would still periodically shut down in the afternoon
hours of the day when it was hottest.
2-11
-------
High Evaporative Emissions Investigation Field Study
Final Report
Unfortunately, the combination of the limited maximum resolution available for these
camera models and the camera-to-pump distance produced images and videos that were not
sufficient to read individual license plates. It is recommended that for future efforts the cameras
be placed on the fueling islands or other locations closer to the test vehicles, or budget for more
advanced cameras that could capture legible license plates at the distances being used for the
task. Adequate cooling and shelter for the devices should also be planned to prevent overheating.
2.5 Weather Station
An AcuRite model 06006 Weather Sensor was used to record key weather conditions
including wind speed, wind direction, and temperature under the gas station canopy. The weather
station, shown in Figure 2-9, was placed beneath the Costco Arvada gas station canopy to
estimate the local conditions experienced by the refueling vehicles and their evaporative plumes.
The weather station receiver logged comma delimited data of the following variables in 12-
minute intervals: temperature, humidity, barometric pressure, rain, wind speed, wind speed
average, peak wind, and wind direction. Because the weather station was placed beneath the gas
station canopy, the rain measurements will not be used.
The weather data - particularly the wind speed and direction - could be used to explain
plume behavior. The receiver was located about 40 meters from the weather sensor. For
independence from Costco line power, the receiver was powered by a lead/acid car battery and
an inverter, as shown in Figure 2-10. The weather data include timestamps written by the
receiver, and the resultant comma-separated values files were individually saved and dated. The
files were then merged with the Costco Arvada gas station dataset via the timestamps with the
SAS program. Because the weather data were recorded in 12-minute intervals, the end-of-
interval readings were applied to each span of time.
2.6 Colorado Vehicle Registration Information
For the purposes of analyzing the refueling data of different vehicles coming through the
station, we wanted to be able to confirm a vehicle's identity at the time they were refueling. The
reason for this was that our experience has been when recording license plates and vehicle
descriptions, analysis or attempts to confirm identity of the vehicles later during analysis was
unreliable. Therefore, for this study we developed a procedure for identifying vehicles as they
were refueling by looking up their license plates in a recent snapshot of the Colorado vehicle
registration database while it was still possible to further examine the vehicle before it drove
away.
2-12
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 2-9, Weather Station Sensor Array
Figure 2-10. Weather Station Receiver, Display, and Logger Module
2-13
-------
High Evaporative Emissions Investigation Field Study
Final Report
Making this possible at the time of vehicle sampling required that we had a method of
using the observed vehicle's license plate to rapidly lookup the year, make, and model of the
vehicle arriving at the gas pump. This was accomplished by having a technician use an iPad
while on the refueling island to record the license plate of a vehicle refueling at that island. Once
entered, the license plate accessed a lookup file created from the Colorado registration database
that could present the vehicle year, make, and model to the technician for visual confirmation.
Accordingly, CDPHE obtained an April 2019 snapshot of the Colorado registration
database and provided it to ERG for pre-processing before the beginning of field testing. ERG
requested and received only registration database variables for license plate, VIN, year, make,
model, and empty vehicle weight. Vehicle owner name and address were not requested so that
the identity of owners was protected. The empty vehicle weight variable was the only
registration database weight variable that was well populated to help determine vehicle class and,
with model year, to determine the ORVR status of each vehicle. After the end of field testing and
during the data analysis phase of the study, CDPHE provided a July 2019 snapshot, which
allowed us to confirm identities of a few more Colorado vehicles that had refueled at the Arvada
gas station during the field data collection period.
The 6,268,768 registration observations in the July 2019 snapshot were read by a SAS
program8, filtered for relevant observations, and output into a CSV data file for use by the iPad
data collection system. After filtering, 4,976,046 registration observations remained for use by
the iPad system. The following observation filters were used: delete missing plates; delete
missing VINs; delete model years older than 1972; delete non-gasoline fuel types; delete trailers,
motorhomes, buses, and special mobile machinery; and for replicate plates, keep only the
observation with the most recent registration date.
In the Colorado registration database snapshot, the make and model fields were
abbreviated. For example, a Jeep Renegade might be listed as make=JEEP model=REN. While
makes were consistently spelled, models were not, and in many cases, values for models were
blank. We wanted to maximize the iPad technician's ability to confirm vehicle identities by
displaying unambiguous, non-abbreviated model descriptions on the iPad, e.g. Renegade not
REN. Therefore, we used the ERG VIN Decoder to decode the VINs in the snapshot, and where
the VIN decoded without error, the SAS program replaced the registration make and model with
the decoded make and model.
8 P:/CDPHE/Regis2019/COreg_find_mk_mod_yr.sas
2-14
-------
High Evaporative Emissions Investigation Field Study
Final Report
For vehicles with out-of-Colorado plates, since we did not have vehicle registration
information for the other 49 states, the iPad allowed for entering the plate number but allowed
entering only that the plate was out-of-state. However, some iPad technicians took the initiative
to write down the states, plates, and vehicle descriptions for such vehicles. During the analysis
phase, we found that mycar.com could be used to look up several of the confirming vehicle
descriptions for out-of-state vehicles using the plate state and plate number as inputs.
2.7 iPad Data Collection System
We expected that the Costco Arvada gas station would be very busy as customers waited
in line and refueled their vehicles. Therefore, even though we were going to record data and
observe refueling emissions at one refueling island at a time, we needed to develop a method of
data collection that was reliable and easy for a technician who was stationed at the island to use.
We also wanted to use procedures that would minimize concerns and questions asked by gas
station customers. To meet these needs, we developed an electronic data collection system.
The Apple iPad was selected as the electronic field device for its ease of use and the high
availability of programming resources and data collection applications already developed for it.
The final designed and implemented data collection system was custom-built on the Microsoft
PowerApps platform and used the corresponding Apple iOS operating system "PowerApps"
application that allows PowerApps programs to run directly on the iPad. The data collection
application presented a graphical user interface for the technician to enter information on the
iPad for each fueling event while right at the gas pump. The graphical interface showed data
entry fields for all pumps simultaneously so that multiple fueling events at different pumps could
be tracked at the same time. Exact data entry steps for the application are detailed further below.
A mobile internet connection on the iPad allowed for the data to be uploaded in real time
as it was collected, which was sent to a database located on a Microsoft SQL Server instance
running on the cloud-based Microsoft Azure platform. A second database running on the same
platform containing around 5 million Colorado vehicle registration records was used by the iPad
application to rapidly look up license plates as the technician entered them in the license plate
field and returned make, model, and year of the vehicle. During the active data collection portion
of the project, the new data in the SQL Server database was exported daily to a flat text file that
was reviewed for quality and preliminary analysis.
Figure 2-11 shows a screen capture of the iPad interface developed for this study. The
screen is made up of three columns that represent the three islands at the gas station. The four
gas pumps at each island are numbered 1 to 12 on the screen so that they correspond to their
2-15
-------
High Evaporative Emissions Investigation Field Study
Final Report
positions on the three islands. The arrangement of the pump numbers on the screen is the same as
the arrangement of the physical pumps on the islands at the gas station. The background color in
the screen indicates the island where data is currently be being collected. In Figure 2-11 the data
is being collected on the center island since its background is black and the two side columns are
"grayed out."
Figure 2-11. Refueling Vehicle Data Collection Interface for iPad
F © 3 F
(*& • (
ozL o:
4 F/
0 De
FA © 7 FA Q) 8
3® 3®
R
A
E
C
J
C
©ljB«J(© © De ©
FA Q 5 FA © 6
3® 3®
R
A
E
C
J
c
(X
) De (
S)
Ar ^ Ar
Qy Cx C Cx
PLATE
PLATE
Each island area on the screen is made up of four sets of data collection displays and
buttons, so that data from the four pumps can be collected simultaneously. Consider the buttons
for Pump 7 at the top left of the center island in Figure 2-11. The technician pushes the top
yellow button labeled "Ar" (Arrive) when a vehicle arrives at the pump. If the vehicle has a
Colorado plate, they push the white and red "CV" button, otherwise they push the gray and black
"C*" button to indicate a non-Colorado plate. Then, the technician uses a keypad that appears on
the iPad screen to enter the vehicle's license plate, which will appear in the white oval labeled
"PLATE." If the plate is a Colorado plate, the iPad application requests a look-up over the
internet connection from the cloud-hosted SQL Server database with Colorado vehicle
2-16
-------
High Evaporative Emissions Investigation Field Study
Final Report
registrations. If the Colorado plate registration can be found, the app displays the make, model,
and year of the vehicle in the black rectangle below the white "PLATE" entry field. The
technician then compares the iPad-displayed make and model with the vehicle that is in front of
them at the gas pump. If they agree, the technician pushes the green "ACC" ("Accept") button. If
they don't agree, he pushes the red "REJ" ("Reject") button. When the vehicle driver has finished
refueling and begins to pull away from the gas pump, the technician pushes the blue "De"
(Depart) button at the bottom of the screen pump display.
If the vehicle has a non-Colorado plate, then the technician enters the plate, but since the
system does not contain registration databases for all states, it cannot display a year, make, and
model.
Additional buttons for each pump area on the screen help the technician enter data for
unusual situations and for special data collection needs. The blue "F" (Forgot) at the top left of
each pump area is pressed by the technician in the event that a new vehicle has arrived at that
pump and the technician had forgotten to press "De" (Depart) for the previous vehicle at that
pump. The orange triangle with an exclamation mark at the bottom right of each screen pump
area forces the data collection for the current vehicle to terminate and the entry fields to reset.
The clock icon on the lower left of each screen pump area is used to put a clock synchronization
timestamp in the database on demand, which was performed periodically by the technicians to
place a benchmark in the database to be used later for synchronization from other data sources,
such as videos or gas pump fueling data received from Costco.
Refueling activity at the gas station island is complex because activities at each pump are
independent of each other. This means that individual vehicles can arrive and depart and pump
fuel at any time regardless of what is occurring at other pumps on the island. Therefore, the iPad
interface was designed so that the technician could collect data on all four pumps at the same
time subject to the constraint that the order of button pushing for a given pump was constrained
to follow a logical sequence. To help the technician determine where each vehicle's data entry
was in the sequence, after buttons are pushed on the iPad screen, button backgrounds change
color to a pink highlight to indicate that they have been pushed.
2-17
-------
High Evaporative Emissions Investigation Field Study
Final Report
Individual data collection steps for observing a refueling event are recorded and uploaded
as they are completed, rather than waiting for an entire fueling event to be completed and that
full record to be uploaded. As buttons on the iPad interface are pushed for each observation step,
a timestamp for each push and the corresponding keystroke event code for that button or
information entered in a field are transmitted via the mobile internet connection to the cloud-
hosted Microsoft SQL Server database. Table 2-3 gives the keystroke event codes. Table 2-4
shows an excerpt of the database to provide an idea of the format of the data being collected.
Table 2-3. iPad Keystroke Event Codes Written
to Microsoft SQL Server Database9
Code
Meaning
A
"Arrive" New vehicle is arriving at the pump, but the plate state has not been entered
yet.
C
Colorado plate.
O
Out of state plate.
R
"Reject" The make/model of the vehicle at the pump does NOT match the
registration database make/model found by looking up the vehicle's license plate in
the database.
T
"Accept" The entered plate is verified (good) against the vehicle appearance. The
make/model of the vehicle at the pump DOES match the registration database
make/model found by looking up the vehicle's license plate in the database.
D
"Depart" Vehicle is leaving the pump. Writes the plate and vehicle registration
lookup results (VIN, year, make, model, empty vehicle weight).
F
"Forgot" Technician did not observe when vehicle left the pump and either now sees
the pump is empty or that a different car is present. Writes the plate and lookup
results to database, just as for "D".
E
"Reset" Start iPad data collection for this refueling event. Pump; plate and vehicle
lookup results are NOT saved or written to database; they are only written on a "D"
or "F".
S
"Sync" Clock synchronization timestamp.
The iPad interface was developed during repeated visits to the Costco Arvada gas station
before the actual field study began. During these visits early prototypes of the iPad interface and
the associated collection system were tested and then modified until the entire system was easy
for the technician to use and could collect accurate vehicle and refueling data.
9 C:\Documents\EPA CanisterDegradation\WA2-23
(GasStnRebellion_MAR2019)\QAPP/DataProcessingSteps_eMail-190711 .msg
2-18
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 2-4. Sample Data in Microsoft SQL Server Database
as Uploaded from the iPad Interface10
dbo GasStnMon2019vl
msEpochTime
Pump
ID
Event
Recorded
PLATE
Year
Make
Model
VIN
DbPlatelD
TimeDateStamp
Empty
Weight
1563936332205
5
D
ABC123*
2016
TOYOTA
Highlander XLE
5TDJKRFH1GS******
ABC123*
7/24/2019 2:45:54 AM
4400
1563936227562
5
T
7/24/2019 2:43:49 AM
1563936207970
5
C
7/24/2019 2:43:29 AM
1563936204659
5
A
7/24/2019 2:43:27 AM
1563936146511
6
D
DEF456*
2002
SUBARU
Impreza
JF1GD29692G******
DEF456*
7/24/2019 2:42:41 AM
3100
1563936017867
4
E
7/24/2019 2:40:20 AM
1563936014003
4
A
7/24/2019 2:40:16 AM
1563935994433
6
T
7/24/2019 2:39:55 AM
1563935971942
6
A
7/24/2019 2:39:37 AM
1563935973132
6
C
7/24/2019 2:39:37 AM
1563935969802
5
E
7/24/2019 2:39:32 AM
1563935968079
5
A
7/24/2019 2:39:29 AM
1563935832402
7
D
GHJ789*
2011
DODGE
Charger R/T
2B3CM5CTXB******
GHJ789*
7/24/2019 2:39:14 AM
4400
1563935772498
5
D
KLM321*
1992
FORD
F150 Regular Cab
1FTDF15H4NK******
KLM321*
7/24/2019 2:36:15 AM
4200
1563935770742
7
T
7/24/2019 2:36:12 AM
1563935723696
7
C
7/24/2019 2:35:25 AM
1563935672015
7
A
7/24/2019 2:34:34 AM
1563935598359
5
T
7/24/2019 2:33:19 AM
1563935567212
5
C
7/24/2019 2:32:48 AM
1563935561897
5
A
7/24/2019 2:32:43 AM
1563935513204
7
D
PQR654*
2002
CHEVROLET
Blazer 4WD
1GNDT13S822******
PQR654*
7/24/2019 2:32:10 AM
4600
1563935366552
7
T
7/24/2019 2:29:28 AM
1563935337335
7
C
7/24/2019 2:28:58 AM
1563935336026
7
A
7/24/2019 2:28:58 AM
1563935325620
5
D
STU987*
2009
JEEP
Patroit LHD4WD
1J4FF28B69D******
STU987*
7/24/2019 2:28:54 AM
3200
1563935251707
5
T
7/24/2019 2:27:33 AM
1563935234319
5
C
7/24/2019 2:27:16 AM
1563935233292
5
A
7/24/2019 2:27:15 AM
1563934944319
7
D
VWX123*
2007
CHEVROLET
1500 4WD
3GNFK12347G******
VWX123*
7/24/2019 2:26:55 AM
5700
2.8 Gas Station Logsheets
As testing began at the Arvada Costco gas station, it became apparent that the refueling
behavior of customers was sometimes unusual. In these special cases, the iPad data collection
interface could not be used to document everything that was happening. Therefore, we began
collecting supplemental information on paper logsheets for potential use during data analysis.
The logsheets had columns for day of the week, date, pump number, nozzle hang-up time, and
comments.
The types of events that produced paper logsheet entries included: multiple vehicles
refueled on a single purchase, refueling of gas cans either by themselves or with a vehicle
refueling, refueling lawn mowers brought to the gas station by a lawn mowing company, break-
down of a vehicle at a fuel pump thereby blocking the pumps for use by others, multiple
refuelings with multiple purchases on one vehicle at a pump, vehicles that entered credit card
information on the pump but did not actually pump any fuel, notes regarding incorrect iPad
button pushes, vehicles that partially fueled at one pump then pulled up and continued fueling at
10 License plate numbers in this table are artificial.
2-19
-------
High Evaporative Emissions Investigation Field Study
Final Report
a second pump, make and model information on non-Colorado license plate vehicles, and license
plates of the additional vehicles refueled on the same credit card purchase.
2.9 Refueling Event Listing
After field data collection was complete, a SAS program combined each refueling event
from the iPad data into one entry and synchronized the information to several other sources of
data that were obtained at the same time, such as weather conditions, fueling volumes and credit-
card timestamps from Costco, and file names of videos of each refueling event captured by the
Rebellion Photonics GCI camera. This synchronizing and merging of multiple data sources
produced a flat data file with a single-line entry for each fueling event.
The Arvada gas station field data collection effort produced several datasets:
• GCI camera 30-second Enhanced MidWave video files,
• iPad keystroke data,
• Weather station data, and
• Gas station pump transaction data.
We merged the above datasets in a way that produced a list of events with a single
observation for each refueling event. We call that listing the event-by-event listing (ExE). After
the ExE was created, we imported it into a SharePoint database so that all videos for a given
refueling event could be viewed and evaluated. Then, the results of the video viewings were also
appended to the SharePoint database.
Merging the datasets and creating the ExE was done by a SAS program11. Because the
different datasets contain information on events as a function of time, we needed to merge them
in the time domain. The problem with time-merging was that events from the different datasets
never occurred at exactly the same time as events from another dataset. So, we could not simply
merge by the time variable. Our solution was to put all data in a time-based listing that we call
the piano roll12. The piano poll construction began with a file that had a time scale with one
observation for each second and for all seconds from the beginning to the end of the field data
11 P:\EPA_RefuelingEmissions_WA2-23\Summer2019\Analysis/read_field.sas
12 Piano roll refers to the roll of paper used to operate the keys on a player piano. As the piano roll moves,
punched holes in the paper tell which and when each piano keyboard key is depressed. A visual
inspection of the piano roll tells when each key should be activated, or in our case, when each dataset
activity was occurring.
2-20
-------
High Evaporative Emissions Investigation Field Study
Final Report
collection period. Except for the Linking DateTime field, all other fields were initially blank. We
created a group of fields for each of the four pumps (A, B, C, D) at an island.
Since the video filenames contained the date and time of the beginning of each video, and
each video was exactly 30 seconds long, the filename was written to the filename field, which is
common to all pump positions at the island; otherwise, the video filename field was left blank.
The iPad keystroke information, which included vehicle description information, was entered for
each pump position variable in a similar fashion since each keystroke had an associated
timestamp. The vehicle descriptions were lagged down the piano roll vehicle fields to reflect
when a vehicle was at the pump. Thus, for one refueling event, the vehicle descriptions at a given
pump were the same from the time when the vehicle arrived until it departed. After that time
period, the vehicle description fields were blank until the next vehicle arrived at the pump.
Similarly, the gas station transaction data was used to enter the gallons dispensed for each pump
position from the credit card approval timestamp through the pump nozzle hang-up timestamp.
The weather station data was added to the piano roll so that all weather fields were filled based
on the weather station's average values for each 12-minute datalogging period. Finally, we
brought in the transcribed reference vehicle test conditions and timing information.
We used the piano roll to check the time-alignment of the different datasets - particularly
the Rebellion video time, the iPad keystroke time, and the Costco transaction time. We found
that all three were already synchronized within 3 seconds. So, we made no adjustments to the
time bases. The one exception was that we found that the transactions from Pump 9 were off by
4 minutes for a period on July 9. We corrected the times for that period.
Once the piano roll was complete, the SAS program used it to create the ExE listing. That
process worked by looking down the time series at the set of variables assigned to each pump
position and retaining values for each refueling event appropriately. For example, a typical
refueling event would start with a timestamp for arrival for one pump position. Then, as the
program worked its way down the piano roll, it would retain credit card approval timestamp,
video filenames for videos taken during the refueling event, nozzle hang-up timestamp, gallons
dispensed, vehicle description, and finally the departure timestamp. We decoded the VINs in the
ExE to provide vehicle types.
The counts of observations as merging proceeded give an indication of how merging and
filtering affects the size of the final set of data to be analyzed. We videoed on one side of each
day's selected island from 9:14 a.m. on July 8 through Jul 23. On the fourteen days that we
videoed, there were 27,689 gas pump transactions. On those days and on both sides of the
2-21
-------
High Evaporative Emissions Investigation Field Study
Final Report
videoed islands, there were 8,729 refuelings. During this period, we recorded 31,487 iPad
keystroke event codes on refuelings on both sides of the selected island and, thereby, collected
iPad information on 7,240 refuelings on both sides of the island. After considering that videos
were only of refueling on one side of the island, we counted 3,817 refueling events where we
have both iPad information and videos. We have 2,895 refueling events with videos on vehicles
with Colorado plates that led to confirmed vehicle descriptions (VIN, year, make, model, and
vehicle type). Of these, 2,376 had clear vehicle type assignments of LDGV, LDGT12, or
LDGT34.
2.10 SharePoint Database
The ExE listing that was created by read_field.sas was read into a Microsoft SharePoint
database using a program developed using the Microsoft Flow platform to automatically create
and populate list entries on a Microsoft SharePoint Online website where team members could
easily review the data, as shown by the example in Figure 2-12.
Figure 2-12. Example SharePoint Refueling Event Selected for Review
ERG Mobile Sources Collaboration Site
4? Edit Share 4b Copy link © Delete c/1 Flow
GasStnMon2019 EPA 01
Title
89X6924DB1N799
58U7145YZ6L380
65L4419TB0Q434
87E1382JC9K591
7O257B135030N3
0Q102Z21Z281F8
99R0245QH2I670
89O5450KK5Y395
1O530C18C117V3
56J2592WS6Y326
63D7970OU2L153
4U502N51H127O6
6Q967E54H380Q8
0I762A44S063R0
Count
EPA QA Review Co...
i_Make
HONDA
JEEP
TOYOTA
CHEVROLET
HONDA
65L4419TB0Q434
EPA QA Review Comments
Enter value here
LVehicleShort
1989_BIIICK_REATTA_3300
i_ShortVehide_Match
Match
LPumpID
7
c_Gallons
14.176
i_Arrive_MTN
08JUL19:15:55:33
r_EnhMW_1_Video
viewerj 562622958208.mp4
•A
^v|
—* i
HfW 1
(~) 00:03 ¦
00:27 © ® ©
2-22
-------
High Evaporative Emissions Investigation Field Study
Final Report
Each entry on the left side of the screen in the list represents a refueling event. By
clicking on the Title in this list, the details of the refueling appear, as shown on the right side of
the screen. In this example, the video in the lower right shows the refueling plume of the rear
vehicle, a 1989 Buick Reatta (pre-ORVR), as a white fog. Each refueling event contained one to
six videos, depending on the amount of time that a vehicle was at a gas pump. As part of the SAS
merging program, the correct video was embedded into each SharePoint list record entry so that
the reviewing technician could easily find and watch the videos for the event and make
observations. During review the technician entered additional data into SharePoint records, such
as the volume of fumes observed in the video during the fueling event. After all the records were
reviewed, the entire SharePoint list of events was exported for importing into SAS to update the
ExE listing to produce a final dataset for detailed analysis.
2-23
-------
High Evaporative Emissions Investigation Field Study
Final Report
3.0 Arvada: Analysis of Refueling Data
We viewed Enhanced MidWave videos for the subset of all refueling events where we
had a complete set of vehicle and plume information. Specifically, the subset of videos that were
viewed had the following criteria: the vehicle had a Colorado license plate, a corresponding
registration database VIN with non-missing model year, make, and model, a refueling event with
at least one Enhanced MidWave video, and the video contained the vehicle that was refueling.
We did not view videos of vehicles with non-Colorado plates, Colorado plates that could not be
found in the April or July 2019 vehicle registration database snapshots, VINs that could not be
decoded without errors or vehicles that were not videoed by the GCI camera.
The Enhanced MidWave videos were viewed in two phases. Initially, in Phase 1 we
viewed all videos for refueling events that were selected as described above. The viewing
instructions are presented in Appendix B. The results of those Phase 1 (preliminary) viewings are
described in Section 3.4. During analysis of the Phase 1 viewings, we realized that short-duration
puffs of emissions were occurring in many refueling events. Most of these puffs seemed to be
associated with refueling activities, such as gas cap removal, start of fuel flow, nozzle click-off at
the end of fuel flow, fuel tank topping-off, and not associated with the bulk fuel flow. We
suspected that under these puff circumstances the as-designed evaporative emissions control
systems of any vehicle may not be able to control the puffs. We judged that the Phase 1 video
viewing results indicated an artificially high level of evaporative emission control system
malfunction in the fleet. Therefore, in Phase 2 we re-viewed the videos of refueling events on
confirmed ORVR vehicles that had any refueling emissions seen in the Phase 1 video
examinations. The Phase 2 viewing instructions are given in Appendix C. The analysis of those
results begins in Section 3.5.
3.1 Characteristics of the Sampled Vehicles
Before we describe the analysis of refueling emissions, this subsection presents
characteristics of the sample of the Denver area vehicle fleet that refueled at the Arvada Costco
gas station where the data was collected. Because the sample is from just a single gas station and
the customers can only be Costco members, we cannot claim that the sample is representative of
the Denver-area fleet; however, trends in the data can be used to explore behaviors and
relationships that may be present in other fleet samples. Here we consider only those vehicles
with refuelings that produced usable GCI videos.
Several different classes of vehicles refueled at the Arvada gas station during the study.
Table 3-1 shows the eight vehicle classes evaluated and their gross vehicle weight rating
3-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
(GVWR) range descriptions. Because the pairs LDGT1 and LDGT2, and LDGT3 and LDGT4
have the same GVWR range descriptions, in this report we refer to them as LDGT12 and
LDGT34.
Table 3-1. Description of Gasoline Vehicle Classes
Vehicle
Class
Description
GVWR Range
(pounds)
I.IXjY
Light-Duty Gasoline Vehicles
Passenger Cars
LDGT1
Light-Duty Gasoline Trucks 1
0 - 6,000
LDGT2
Light-Duty Gasoline Trucks 2
0 - 6,000
LDGT3
Light-Duty Gasoline Trucks 3
6,001 - 8,500
LDGT4
Light-Duty Gasoline Trucks 4
6,001 - 8,500
HDGV2B
Class 2b Heavy-Duty Gasoline Vehicles
8,501 - 10,000
HDGV3
Class 3 Heavy-Duty Gasoline Vehicles
10,001 - 14,000
HDGV4
Class 4 Heavy-Duty Gasoline Vehicles
14,001 - 16,000
The presence or absence of an ORVR system on a vehicle can be expected to have a large
effect on the size of refueling emissions. Since the on-site technician used the iPad app, the
license plate, and the Colorado registration database to visually confirm the identity of the
Colorado vehicles at the Arvada gas station, we can determine the ORVR equipment in the
sampled fleet. We used the VIN in the registration database to determine the gasoline vehicle
class and model year for each vehicle. We used that information with the ORVR implementation
schedule to determine ORVR equipment.
ORVR implementation schedules depend on vehicle class and model year. We have
evaluated these vehicle classes in the sample: LDGV, LDGT1, LDGT2, LDGT3, LDGT4, and
HDGV2b. We did not evaluate sampled vehicles in heavier vehicle classes since their ORVR
implementations are complex and there were only about 16 of those vehicles anyway.
In Table 3-2 we group vehicle classes that have the same ORVR implementation
schedule. As an example, for LDGVs, all model years before 1998 had no ORVR equipment,
and all model years 2000 and after did have ORVR. 1998 and 1999 were ORVR implementation
transition model years for LDGVs. For the transition model years, we have just estimated the
number of non-ORVR and ORVR counts by applying the minimum phase-in percentage that
manufacturers had to follow to the vehicle count in each transition model year. Since the number
of sampled vehicles in the transition years is a relatively small fraction of the total number of
vehicles in the sample, estimating the ORVR/non-ORVR apportionment for transition model
3-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
years contributes only a small uncertainty. The table shows that about 90% of the Costco Arvada
customers that we sampled during the three-week study in July 2019 were originally equipped
with ORVR systems.
Table 3-2. ORVR Equipment in the Arvada Sample13
Vehicle
Class
(GVWR Range)
Model
Year
Group
ORVR
Implementation
Vehicle
Count
non-
ORVR
ORVR
LDGV
(passenger cars)
pre-1998
0%
23
23
1998
40% min
9
est 5
est 4
1999
80% min
17
est 3
est 14
2000-2019
100%
723
723
LDGT1 and LDGT2
(0 - 6000 lbs)
pre-2001
0%
78
78
2001
40% min
25
est 15
est 10
2002
80% min
30
est 6
est 24
2003-2019
100%
1046
1046
LDGT3 and LDGT4
(6,001 - 8500 lbs)
HDGV2B
(8,501 - 10,000 lbs)
pre-2004
0%
108
108
2004
40% min
27
est 16
est 11
2005
80% min
25
est 5
est 20
2006-2019
100%
490
490
Total 2601 259 2342
(%) 100% 10.0% 90.0%
Figure 3-1 through Figure 3-4 show distributions of vehicle model year, empty weight,
tank capacity, and canister capacity. The refueling vehicles generally had newer model years,
shown in Figure 3-1. Less than 10% of vehicles were model year 2001 or older, and about half of
vehicles were model year 2012 or newer. We used the Colorado vehicle registration database and
the vehicle's license plate to look up the vehicle empty weight, which was the only weight
variable widely populated in the database. Figure 3-2 shows the distribution of empty vehicle
weights. EPA was able to determine the fuel tank capacities (gallons) and evaporative emission
control system canister capacities (g) for most of the videoed vehicles in the study, as shown in
Figure 3-3 and Figure 3-4. For some trucks, different options for multiple fuel tanks were offered
by manufacturers. In such cases, the tank and canister capacities could not be determined from
generic YIN information.
13 C:\Documents\EPA CanisterDegradation\WA2-23 (GasStnRebellion_MAR2019)\Data
QC/EventByEvent_200225_ORVRdistribution.xlsx
3-3
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-1. Model Year Distribution of Sampled Vehicles
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
201
201
201:
201
201
201
2011
201'
2018
2019
2020
FRE(
CUM.
FREQ.
CUM.
PA.
100
200
300
FREQUENCY
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_ORVR.sas 27FEB20 17:22
3-4
-------
High Evaporative Emissions Investigation Field Study Final Report
Figure 3-2. Empty Weight Distribution of Sampled Vehicles
.c
o>
'55
Q.
E
LU
0
250
500
750
1000
1250
1500
1750
2000
2250
2500
2750
3000
3250
3500
3750
4000
4250
4500
4750
5000
5250
5500
5750
6000
6250
6500
6750
7000
7250
7500
7750
8000
8250
8500
8750
9000
9250
9500
9750
10000
CUM. CUM
FREQ, FREQ, RCT, RCT,
200 300 400
FREQUENCY
600
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumos_ORVR.sas 27FEB20 17:22
3-5
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-3. Fuel Tank Capacity of Sampled Vehicles
CUM.
CUM.
FRE(%
FRE4)
po%
po%
33
33
1.35
1.35
68
101
2.78
4.12
136
237
5.55
9.68
65
302
2.65
12.33
183
485
7.47
19.80
271
756
11.07
30.87
225
981
9.19
40.06
125
1106
5.10
45.16
422
1528
17.23
62.39
197
1725
8.04
70.44
222
1947
9.06
79.50
66
2013
2.69
82.20
114
2127
4.65
86.85
6
2133
0.24
87.10
63
2196
2.57
89.67
109
2305
4.45
94.12
3
2308
0.12
94.24
34
2342
1.39
95.63
0
2342
0.00
95.63
12
2354
0.49
96.12
22
2376
0.90
97.02
3
2379
0.12
97.14
3
2382
0.12
97.26
2
2384
0.08
97.35
20
2404
0.82
98.16
30
2434
1.22
99.39
0
2434
0.00
99.39
13
2447
0.53
99.92
1
2448
0.04
99.96
0
2448
0.00
99.96
0
2448
0.00
99.96
0
2448
0.00
99.96
0
2448
0.00
99.96
0
2448
0.00
99.96
1
2449
0.04
100.00
100
200
300
400
500
FREQUENCY
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_ORVR.sas 27FEB20 17:22
3-6
-------
High Evaporative Emissions Investigation Field Study Final Report
Figure 3-4. Canister Capacity Distribution of Sampled Vehicles
CUM.
3
'c
ro
U
PCX
0.00
0.00
0.00
0.49
0.00
0.12
0.33
0.04
0.16
1.31
3.31
5.19
12.56
19.88
10.84
8.55
6.91
12.07
3.11
8.79
1.19
0.25
1.51
0.37
0.53
0.04
0.20
2.00
0.25
0.00
0.00
CUM.
PCX
0.00
0.00
0.00
0.49
0.49
0.61
0.94
0.98
1.15
2.45
5.77
10.96
23.52
43.39
54.23
62.78
69.69
81.76
84.87
93.66
94.85
95.09
96.61
96.97
97.51
97.55
97.75
99.75
100.00
100.00
100.00
100
200
300
400
500
FREQUENCY
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_ORVR.sas 27FEB20 17:22
3-7
-------
High Evaporative Emissions Investigation Field Study
Final Report
The median gasoline tank capacity was 19 gallons with a range of 11 to 38 gallons. The
median canister capacity was 140 grams with a range of 30 to 275 grams. We expected that
vehicles with larger gas tanks would be fitted with larger capacity canisters since larger gas tanks
impose a larger demand on the evaporative emission control system. The burden is particularly
large for ORVR vehicles since the control system must limit refueling emissions. Therefore,
Figure 3-5 shows a plot of canister capacity against tank capacity for LDGV, LDGT12, and
LDGT34 vehicles with confirmed ORVR evap systems. The slope of the linear trend of canister
capacity with tank capacity is about 6.2 grams/gallon.
3-8
-------
High Evaporative Emissions Investigation Field Study Final Report
Figure 3-5. Canister Size vs. Fuel Tank Capacity for Confirmed ORVR Vehicles
300
275
250
225
B 200
N 175
<75
S- 150
0}
£ 125
£
rc
U 100
75
50
25
0
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_ORVR.sas 27FEB20 17:22
. t +
"H- +
4+
+4f
+ H* -
^ + i- ±+ # II IH-TI + H- +
f- + + ' ± + + -H-
-4- ±H I" TI
4- =fcilr ~444- 4-4-4- 4-
# +
ti ¦ ++
+++
+#+ + +
+ +
10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46
Tank Capacity (gal)
3-9
-------
High Evaporative Emissions Investigation Field Study
Final Report
For most refueling events, we were able to calculate the refueling percentage as the ratio
of the gasoline dispensed (from the Costco transaction data) and the tank capacity (from the EPA
look-ups). Figure 3-6 shows the refueling percentage as a function of model year. Events with
percentages over 100% can arise when customers fuel gas cans or non-road vehicles, such as jet
skis or lawnmowers, or when tank capacity records are inaccurate. The distribution of refueling
percentages is shown in Figure 3-7. The median refueling percentage was 71%. About 6% of
customers pumped 90% or more of their vehicle's tank capacity.
3-10
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-6. Percent Refueling vs. Model Year for Sampled Vehicles
140
130
120
Co 110-
~ 100-
r? 90
•+-*
S 80
u
aJ 70
% «
•J 50
Qi
= 40
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-7. Distribution of Refueling Percentage for Sampled Vehicles
0)
u\
ro
¦*->
£
-------
High Evaporative Emissions Investigation Field Study
Final Report
3.2 Analysis of Reference Vehicle Artificial Refueling Emissions Releases
The metered releases of artificial refueling emissions from the reference test vehicle were
described in Section 2.2. For 30 seconds, mixtures of butane or gasoline headspace vapor were
released at a rate of 10 gallons/minute from either the vehicle's left fuel fill door (LDOOR), an
imaginary right fuel fill door (RDOOR), or from on top of the fuel tank (TANK). These releases
produced continuous plumes - not puffs as will be observed and discussed later during gas
station customer refuelings. The mixtures had nominal concentrations of 10%, 30%, or 100%
relative to the equilibrium headspace HC vapor concentration. The 100% relative concentration
was a 50% molar concentration in nitrogen. For butane, the 100% relative HC concentration was
4.5 g butane/gallon of mixture vapor14. For gasoline headspace vapor, the 100% relative HC
concentration was 4.6 g HC/gallon of headspace vapor (see Appendix E for the estimate.).
The test conditions and results of viewing the Enhanced MidWave videos are tabulated in
Appendix A. A summary of those results as a function of HC vapor type (butane, gasoline
headspace vapor), release location, and relative HC concentration in the release mixture is shown
in Table 3-3. The denominator of the ratio within each cell of the table gives the number of runs
at the test condition, and the numerator gives the number of runs that had an observable plume in
the video. Of the 108 valid releases that were successfully videoed, 94 had observable plumes.
Table 3-3. Video Plume Visibility Responses to Test Vehicle Conditions15
Relative HC Concentration
HC Type
Release
Location
10%
30%
100%
Butane
(BUT)
LDOOR
5/6
6/6
6/6
RDOOR
5/6
7/7
7/7
TANK
2/6
7/7
6/6
Gasoline
Headspace
Vapor
(GAS)
LDOOR
4/6
6/6
5/5
RDOOR
6/6
6/6
5/5
TANK
1/6
5/6
5/5
94/108
14 (0.5 ft3 butane/ft3 mixture) * (28.3 L/ft3 butane) * (1 mole butane/22.4 L butane STP) * (492°R/530°R)
* (58 g butane/mole butane) * (1 fit3 mixture/7.48 gal mixture) = 4.5 g butane/gal mixture
15 C:\Documents\EPA CanisterDegradation\WA2-23
(GasStnRebellion_MAR2019)\Report_Final/RefVehicleCounts.xlsx
3-13
-------
High Evaporative Emissions Investigation Field Study
Final Report
The green cells in Table 3-3 indicate the test conditions where all of the runs had
observable plumes. The yellow cells indicate the test conditions where most, but not all, runs had
observable plumes. The pink cells indicate the test conditions where less than half of the runs
had observable plumes. The clearest trend in the table is that as the relative HC concentration
decreases the chances of observing a plume in the video decreases. Also, the results for 10%
relative HC concentration demonstrate that the chances of observing a plume are about the same
for release from the left door and the right door, but releases from under the rear of the vehicle
on top of the gas tank are less likely seen in the Enhanced MidWave videos.
We wanted to use the reference vehicle test results to quantify the ability of the GCI
camera and observations of plumes in the Enhanced MidWave videos. We used logistic
regression to explore the influences of relative HC concentration, refueling emissions release
location (left door, right door, top of fuel tank), release HC type (butane, headspace vapor), fuel
pump location (front, back), and air movement (calm, non-calm) on plume visibility. For
modeling purposes, we used the natural log of the relative HC concentration and defined calm air
movement when the measured wind speed was less than or equal to 1.3 mph. Release location,
release HC type, pump location, and air movement were categorical variables in the regression.
Plume visibility was the logistic regression response variable: 1 = a plume was observed in the
Enhanced MidWave video, or 0 = no plume was observed.
After exploring several logistic regressions16, the best regression described plume
visibility as depending on relative HC concentration and release location with strongly
significant coefficients. The other variables had no significant influence. The predicted
probabilities (fractions) and their 95% confidence intervals for the test conditions are given in
Table 3-4 and compare well with the counts in Table 3-3.
16 P:\EPA_RefuelingEmissions_WA2-23\Summer2019\Analysis/find_plumes_RefVeh.sas
3-14
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-4. Video Plume Visibility Probabilities for Test Vehicle Conditions17
Relative HC Concentration
HC Type
Release
Location
10%
30%
100%
Butane
(BUT)
LDOOR
0.919
0.757
0.460
0.999
0.993
0.916
1.000
0.999
0.986
RDOOR
0.989
0.918
0.595
0.999
0.998
0.967
1.000
0.999
0.995
TANK
0.535
0.239
0.079
0.990
0.934
0.656
1.000
0.999
0.925
Gasoline
Headspace
Vapor
(GAS)
LDOOR
0.919
0.757
0.460
0.999
0.993
0.916
1.000
0.999
0.986
RDOOR
0.989
0.918
0.595
0.999
0.998
0.967
1.000
0.999
0.995
TANK
0.535
0.239
0.079
0.990
0.934
0.656
1.000
0.999
0.925
The logistic regression model can also predict plume visibilities at relative HC
concentrations different from those tested with the reference vehicle releases. We used the
regression model developed from the reference vehicle data to calculate the plume visibility
probabilities for the three release locations across the full range of relative HC concentrations.
The results are shown in Figure 3-8. The figure has two x-axes to indicate the relative HC
concentrations (%) and the estimated refueling emissions concentrations (g/gal) using 4.6 g/gal
at 100% as the basis. The concentrations with a 50% probability of observing refueling
emissions for the three release locations were 0.2, 0.3, and 0.7 gHC/gallon, respectively.
In Figure 3-8, the curves for the right and left door release locations are to the left of the
curve for the tank release location. This means that plumes from releases from the right and left
doors can be seen to lower relative HC concentrations than plumes from releases on top of the
tank under the rear of the vehicle. This makes sense because refueling emissions from release
points under a vehicle disperse to a greater extent before they can be videoed by the GCI camera.
The curves in Figure 3-8 can be used to roughly classify the refueling emission rate of
private vehicles. Consider the curve (blue) for the right door. The blue curve has a "wall" in the
17 C:\Documents\EPA CanisterDegradation\WA2-23
(GasStnRebellion_MAR2019)\Report_Final/RefVehicleCounts.xlsx
3-15
-------
High Evaporative Emissions Investigation Field Study
Final Report
3% to 9% relative HC concentration range, where the plume visibility probability increases
rapidly from 10% to 90%. This wall can be used to separate plumes into low emissions and high
emissions behavior. For example, suppose the video of a vehicle's refueling event shows a plume
coming from the fuel fill door of a vehicle. The blue curve indicates that it is likely that the
relative HC concentration of the emissions in the plume at the point in the refueling event that
the plume is between 5%, which is the 50% probability value of the blue curve, and 100%, which
is the relative concentration of uncontrolled fuel tank headspace vapor. On the other hand, at
times in videos when no plume is observable from the fuel fill door, the blue curve indicates that
the refueling emission rate would be low - likely between 0% and 5% relative HC concentration.
Artificial Evap HC Relative Concentration (%)
0.00 0.23 0.46 0.69 0.92 1.15 1.38 1.61 1.84 2.07 2.30
Estimated Refueling Emissions Concentration (g/gal)
PLOT — Left Door Right Door — Tank
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_RefVeh.sas 02MAR20 13:21
Sometimes, viewing a video can suggest that the emissions location is near the fuel fill
door, but in other cases the source of the refueling emissions plume cannot be determined.
Vehicle manufacturers have varied the location of the canister vent for maintaining functionality
of the canister. Therefore, we do not necessarily know which curve in Figure 3-8 should be used
3-16
-------
High Evaporative Emissions Investigation Field Study
Final Report
to classify the plume emission rate. If the source location is unknown, then all three curves could
be used to define a "fuzzy wall" that extends from about 3% to 25% relative HC concentration.
The predicted probability curves can also be used to estimate the GCI camera detection
limit for plume detection by observing Enhanced MidWave videos. In the simplest terms, and as
described in 40 CFR Part 136 App B, the minimum detection limit (MDL) is a statistical
estimate of the lowest concentration at which there is a 99% chance that the concentration is
greater than zero. Figure 3-9 shows a zoomed-in version of Figure 3-8. The thin, black,
horizontal reference line at 0.99 indicates that the detection limits of the right door, left door, and
tank release locations are 19%, 27%, and 53% relative HC concentration, respectively. Using 4.6
gHC/gallon of vapor as the 100% relative HC concentration,18 these three detection limits
correspond to 0.9, 1.2, and 2.4 gHC/gallon of vapor, respectively. The solid, thick curves give
the best estimate of the plume visibility probabilities, and the thin, dashed curves give the 95%
confidence intervals. These widely spaced confidence interval pairs of dashed lines in the figure
indicate uncertainty in these detection limit estimates.
To put these values in perspective, consider the current applicable refueling standard of
0.2 gHC/gallon, and the estimated concentration of uncontrolled refueling emissions of 4.6
gHC/gallon, which is equal to the vehicle fuel tank headspace hydrocarbon vapor concentration
for the average testing conditions. The detection limits and 50% probability reference values fall
between the 0.2 gHC/gallon standard and the 4.6 gHC/gallon uncontrolled emissions value.
Therefore, in this study, the GCI camera videos can image refueling emissions from evaporative
emission control systems that have no control and probably systems with moderate control, but
systems that have very good control will probably produce refueling videos with no observable
plume. This is good emissions detection behavior for the study since, in general, the videos will
be able to distinguish between control systems with good and poor behavior.
18 See Section 2.2 and Appendix E for using ReddyEvap 2010 to estimate headspace HC vapor
concentration at the field conditions in this study.
3-17
-------
High Evaporative Emissions Investigation Field Study
Final Report
0 10 20 30 40 50 60 70 80 90 100
Artificial Evap HC Relative Concentration (%)
0.00 0.46 0.92 1.38 1.84 2.30 2.76 3.22 3.68 4.14 4.60
Estimated Refueling Emissions Concentration (g/gal)
PLOT -- Left Door Right Door -- Tank
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_RefVeh.sas 02MAR20 13:21
Figure 3-9. Determination of Detection Limit by Release Location
3.3 Potential Effect of Open Driver's Door on Plume Observations
Field data collection, Enhanced MidWave video viewing, and data analysis personnel all
observed that a substantial number of customers left their door open during refueling at the
Costco Arvada gas station. The concern was that open doors could completely obscure otherwise
observable plumes from being seen by the GCI camera.
If doors obscured plumes, then we would expect to see higher plume rates for events with
closed doors. To determine if open doors hid plumes, we randomly selected refueling events of
2007+ model year vehicles with observed plumes and without observed plumes and re-viewed
their Enhanced MidWave videos to determine if the door was open or closed. Table 3-5 shows
the results of the analysis. For the 2007+ model years, 1626 events had no observable plumes,
329 events had light plumes, and 121 had heavy plumes. We viewed videos of 100 of the no-
plume events, of 33 of the light-plume events, and of all 121 heavy-plume events. To make a
3-18
-------
High Evaporative Emissions Investigation Field Study
Final Report
distinction between doors open just for drivers to get in and out of the vehicle and extended
door-open durations, we considered a door open if it stayed open for at least 5 seconds.
Table 3-5. Estimated Effect of Open vs. Closed Door on Plume Observability
Type of Event
Total Events
Sampling
Method
Sampled
Events
Door Open >
5s
% Open ±
95% CLM
no plume
1626
random
100
37
37% ± 10%
Light plume
329
random
33
13
39% ± 17%
Heavy plume
121
all
121
51
42% ± 9%
The table shows that for the three types of plume events, the rate of doors open was
always around 40%. The 95% confidence limits were estimated using pq/N as the estimate of the
variance. We would expect a higher percent-open rate for the no-plume events than for the plume
events, if open doors hid plumes. Since 37% is not larger than 39% and 42% - at the least not
within the uncertainty, we conclude that open doors probably do not greatly affect the ability to
observe plumes. The data indicates that there is no reason to believe that the different types of
plumes had different door-open fractions.
3.4 Model Year Trends in Phase 1 Plume Observations
We watched all videos of each selected refueling event using the Phase 1 viewing
instructions given in Appendix B and judged the Phase 1 plume status: a) no plume, b) light-
density plume, c) heavy-density plume, or d) gasoline puddle with plume after the vehicle left
and were given codes 0, L, H, and P, respectively. The definitions of light-density vs. heavy-
density plumes were arbitrary and left to the video observer to judge. The plumes in the
Enhanced MidWave videos of the reference vehicle were used as a guide. The plume status was
determined without regard to the duration of the plume or when a plume occurred in the video or
in the refueling event. We did attempt to target Phase 1 plume status observations between the
customer's credit card validation at the gas pump and the fuel nozzle hang-up at the end of each
refueling event - if those activities could be determined from the video images. Sometimes it
was difficult to attribute a plume to the front or rear vehicle at the island.
The Phase 1 plume status results for the 2,854 selected refueling events are given in
Table 3-6. The model year trend of the fraction of refuelings with observable plumes, which is
the last column in Table 3-6 is shown in Figure 3-10. The preliminary trend is characterized by
observable refueling plume fractions near 100% for pre-ORVR vehicles (pre-1998) and
moderately low fractions for the newest (2007+) vehicles with a downward trend during the 1998
through 2006 model years.
3-19
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-6. Phase 1 Status of Refueling Events that Met Selection Criteria19
Model
Year
Refueling Event Plume Stai
tus
Total
Events
Events with
Plumes
(fraction)
No
Plume
Light
Plume
Heavy
Plume
Puddle
Plume
1978
0
0
1
0
1
1.00
1983
0
1
0
0
1
1.00
1985
1
0
0
0
1
0.00
1986
0
0
1
0
1
1.00
1988
0
0
1
0
1
1.00
1989
1
0
5
0
6
0.83
1990
1
0
3
0
4
0.75
1991
0
0
2
0
2
1.00
1992
1
2
6
0
9
0.89
1993
2
0
8
0
10
0.80
1994
0
0
9
0
9
1.00
1995
1
0
8
0
9
0.89
1996
2
1
8
0
11
0.82
1997
3
1
13
0
17
0.82
1998
3
3
19
0
25
0.88
1999
14
5
26
0
45
0.69
2000
16
7
40
0
63
0.75
2001
24
8
28
0
60
0.60
2002
38
12
27
0
77
0.51
2003
47
16
21
0
84
0.44
2004
72
16
14
1
103
0.30
2005
86
25
12
0
123
0.30
2006
85
19
11
1
116
0.27
2007
119
31
14
0
164
0.27
2008
101
22
13
0
136
0.26
2009
76
13
4
0
93
0.18
2010
88
25
6
2
121
0.27
2011
118
18
7
0
143
0.17
2012
132
23
15
0
170
0.22
2013
147
29
5
0
181
0.19
2014
155
20
13
0
188
0.18
2015
173
39
12
0
224
0.23
2016
133
37
10
1
181
0.27
2017
174
39
8
0
221
0.21
2018
163
25
7
0
195
0.16
2019
47
8
4
0
59
0.20
Total
2023
445
381
5
2854
0.29
19 C:\Users\TDeFries\Documents\EPA CanisterDegradation\WA2-23 (GasStnRebellion_MAR2019)\
Report_Final\PlumeVideoCounts.xlsx
3-20
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-10. Phase 1 Model Year Trend of Observable Plumes, Puffs, and Puddles
1975 1980 1985 1990 1995 2000 2005 2010 2015 2020
Model Year
/projiyEPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_ORVR.sas 27FEB20 17:22
The table shows that only 3% (=82/2854) of the events were for refueling on pre-1998
vehicles. For these vehicles, which are unambiguously pre-ORVR, 85% (=70/82) of refuelings
had observable plumes in the Enhanced MidWave videos. For comparison, since none of these
vehicles had ORVR systems, the expected observable plume rate would be 100%. The cause of
the discrepancy between 85% and the expected 100% is unknown. As discussed later in this
report, several site factors (pump position, wind, background infrared illumination, obstacles in
the camera's line of site, movement of people and vehicles) influence the visibility of emissions
in the videos. Until then, we note that the rate for the front row of pumps was 83% (=35/42), and
the rate for the back row of pumps was 87% (=35/40).
For the newest (2007+) model year vehicles, 22% (=450/2076) of refuelings had Phase 1
observable plumes in the Enhanced MidWave videos. Additionally, Figure 3-10 shows only a
weak decrease in the plume rate from 2007 to 2019. We found it hard to believe that 20% of
almost brand new 2018 and 2019 vehicles would produce refueling emission plumes unless
something unexpected or unusual was occurring with vehicle pre-refueling driving behavior,
with GCI camera sensitivity, or with Enhanced MidWave video viewing.
3-21
-------
High Evaporative Emissions Investigation Field Study
Final Report
Therefore, we examined factors that we hypothesized could affect the plume observation
rate: vehicle class (LDGV, LDGT12, LDGT34), vehicle make, gallons of fuel dispensed,
refueling time of day, wind speed, ambient temperature, distance between fuel pump and fuel fill
door, and fuel pump number. We were specifically looking for a variable that had a large
influence on Phase 1 values of the plume observation rate for the newest (2007+) vehicles.
Except for the differing implementation transition years (LDGV: 1998-1999, LDGT12:
2001-2002, LDGT34: 2004-2005), the Phase 1 plume observation rates for the three vehicle
classes were quite similar. The model year trends for the four most common makes in the dataset
(Toyota: 699 observations, Honda: 329 observations, Ford: 310 observations, Chevrolet: 207
observations) were quite similar. Whether the amount of fuel dispensed was larger or smaller
than the median 12.5 gallons had no significant effect on the Phase 1 plume observation rate.
Similarly, time of day (6am to 10am, 10am to 5pm, after 5pm), wind speed (less than 3.4 mph
median, greater than 3.4 mph median), ambient temperature (less than 85.5 F median, greater
than 85.5 F median), and fuel pump to fuel-fill door distance (near, far) had no significant effect
on the Phase 1 plume observation rate.
We also thought that it was possible that one fuel pump nozzle might be consistently
leaking gasoline liquid or vapor. Therefore, we looked at the model year trends for the six fuel
pumps used in the study. We did not find any evidence of a leaking nozzle. However, we found
that all three pumps (7, 8, 11) in the back row, i.e. farthest from the GCI camera, had a higher
rate of observed plumes than the three pumps (5, 6, 9) in the front row, i.e. closer to the GCI
camera. Figure 3-11 and Figure 3-12 show the model year trends of Phase 1 plume observations
for the 2,076 refueling events plotted in Figure 3-10 divided into the 983 refuelings at the back
pumps and the 1,093 refuelings at the front pumps, respectively. The difference in plume rates is
seen most clearly in Figure 3-11 and Figure 3-12 for the 2007+ vehicles. For refuelings of 2007+
model year vehicles, the average Phase 1 observed plume rate was 13% ± 2% (=137/1093) at the
front pumps and 32% ±3% (=313/983) at the back pumps; where the uncertainties give the 95%
confidence intervals. Thus, the plume observation rates of refueling at the front and back pumps
are statistically different.
3-22
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-11. Phase 1 Model Year Trend of Front Pump (5, 6, 9) Plumes
j2 1.0
w 0 9
(/) u,y
LU 0.8
a>
-Q
£
$ 0.6
n
O
.c
0.7
0.5
£ 04
c
® 0.3
LU
0 0.2
:§ 0.1
u
ro
1 o.o
•—•
>>>»
1975
.• - •
1980 1985 1990 1995 2000 2005 2010 2015 2020
Model Year
/proj1/EPA_RcfuelingEmi55ions_WA2-23/Summcr2019/Analysis/find_plumcs_ORVR.sas 27FEB20 17:22
Figure 3-12. Phase 1 Model Year Trend of Rear Pump (7, 8, 11) Plumes
1975 1980 1985 1990 1995 2000 2005 2010 2015 2020
Model Year
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_ORVR.sas 27FEB20 17:22
3-23
-------
High Evaporative Emissions Investigation Field Study
Final Report
We further refined the analysis of front vs. back pumps by restricting the vehicle dataset
by using refuelings only for 2007-2018 model year vehicles with vehicle classes unambiguously
assigned as LDGV, LDGT12, and LDGT34. This eliminated HDGV2b class vehicles, vehicles
for which vehicle classes were missing when they were not easily available, and medium-duty
vehicles. This resulted in the dataset dropping from 2,076 refuelings to 1,692 refuelings with 890
refuelings at the front pumps and 802 at the back pumps. Of the 890 refuelings at the front
pumps, 87 had light plumes and 26 had heavy plumes for a plume observation rate of 12.7% ±
2.2% (=113/890). Of the 802 refuelings at the back pumps, 179 had light plumes and 64 had
heavy plumes for a plume observation rate of 30.3% ± 3.2% (=243/802). Thus, there was still a
significant difference in Phase 1 plume observation rates between the front and back pumps.
Since this front pump vs. back pump difference is not likely caused by three leaking back
pump nozzles or high emissions vehicles refueling preferentially at the back pumps, we
suspected that a bias in the Phase 1 viewing of the Enhanced MidWave videos was somehow
occurring.
To gain insight into possible reasons for the difference in Phase 1 observations of plumes
for front pumps and back pumps, we examined the videos of selected refueling events for 2007-
2019 model year vehicles shown in Table 3-6. We looked at a random 30 (10%) of the 329
refueling events designated as Light Plume in Table 3-6. We also looked at the videos for all 121
refueling events that were assigned as Heavy Plume or Puddle in Table 3-6. We saw that plume
duration was correlated with the light and heavy plume assignments. Specifically, all 30 of the
light plumes also had short durations (less than 20 seconds), while many of the heavy plumes
had long durations (more than 20 seconds).
Also, it appeared that the shadow of the gas station canopy on the pavement and the
complexity of the background had an influence on plume assignments. If the pavement behind
the vehicle was shaded by the canopy, then the GCI camera was less likely to make a plume
observable. The back-pump row was more likely to have an illuminated pavement background
since the pavement behind the rear pump vehicles was not under the canopy. On the other hand,
the pavement behind front pump vehicles is always under the canopy and therefore more often in
a shadow. Additionally, the front pump vehicles typically had vehicles refueling behind them.
This caused the background of front pump vehicles to be more complex. We believe that the
decreased illumination and increased complexity of backgrounds for front pump vehicles may be
responsible for the lower rate of Phase 1 plume assignments of front pump vehicles relative to
back pump vehicles - especially if plumes had a short duration.
3-24
-------
High Evaporative Emissions Investigation Field Study
Final Report
Based on that video investigation, we again re-examined the videos of the sample of the
dataset of 1,692 refueling events (that is, the 802 back-pump refuelings and the 890 front-pump
refuelings, described above) with an eye toward plume duration as well as front vs. back pump
row. We used 20 seconds as a demarcation between short- and long-duration plumes. Because
we had seen no long-duration plumes in the 30 light plumes that were sampled, we examined
only the heavy plumes and thus presumed (at this point) that there would likely be no long-
duration light plumes in the dataset.
The results of this long-duration, heavy-plume re-examination indicated no bias between
the front and back pumps. Specifically, of the 890 front-pump refuelings, 13 (1.5%) had heavy
plumes with durations longer than 20 seconds. Of the 801 back-pump refuelings, 12 (1.5%) had
heavy plumes with durations longer than 20 seconds. Thus, we see that by considering the
duration of the refueling emissions, the difference in the Phase 1 plume observation rate between
the front and back pumps has gone away. We believe that this may be because long duration
plumes may be more likely to be seen in the videos regardless of background complexity.
During the re-examination of the 1,692 refuelings, we postulated that refuelings might be
divided into three Phase 1 (preliminary) categories:
• Category 1) 19.0% (=1336/1692) had no observable plumes in the videos,
• Category 2) 19.5% (=331/1692) were estimated to have light or heavy plumes of
short duration (puffs) usually occurring at standard refueling activities (gas cap
removal, the very beginning of fuel flow, the end of fuel flow when the nozzle
clicked off, and/or when the pump nozzle was being carried to or from the
vehicle's fuel fill door), and
• Category 3) 1.5% (=25/1692) had heavy plumes lasting at least 20 seconds.
Category 1 events are of no concern since no refueling emissions were seen. Category 2
events are of some concern since emissions were observed. However, the Category 2 emission
episodes were brief, probably resulted in a low mass of emissions, and the episodes occurred
during activities when the ORVR system could not control the emissions. Category 3 events are
of most concern because they were long duration and occurred while fuel was being pumped,
may have produced larger masses of refueling emissions, and the ORVR system should have
been able to control the emissions.
3.5 Phase 2 Re-Viewing Refueling Videos for Time Trends of Refueling Plumes
We needed to re-view the Enhanced MidWave videos of each of the events to determine
the time trends of plumes seen in the videos. The reason is that in many cases, brief plumes,
3-25
-------
High Evaporative Emissions Investigation Field Study
Final Report
which we call "puffs," appeared when the gas cap was removed, when fuel flow started, when
the nozzle clicked off, or when customers used extra clicks to "top off' their fuel tanks. We want
to distinguish instances of that behavior from the behavior when continuous plumes are being
produced during steady fuel-flow periods.
Refueling emissions appear in the videos as swirling white fog. Figure 3-13 shows a
frame from a video that contains a topping-off puff. The puff is just above the rear vehicle's
driver's left hand. Figure 3-14 shows a still from a video that contains a continuous plume from
the front vehicle. The plume is visible to the right of the driver's side mirror.
The analysis of the data in the previous section indicated that refuelings might be able to
be described using three categories that are defined by plume duration and standard refueling
activities. Since the Phase 1 plume observations of the videos did not consider plume duration
and refueling activities, we decided to re-view the refueling videos of confirmed ORVR vehicles.
We also wanted to evaluate vehicles with model years earlier than 2007, where possible and
convenient, to see trends on older ORVR vehicles. This would allow us to better evaluate the
postulated 3-category classification scheme presented above.
For this recoding of plume information, we considered only vehicles that had confirmed
ORVR systems and only LDGVs, LDT12s, and LDT34s. Additionally, we considered only
vehicles with Colorado plates and those where we could find VIN, model year, make, and model
in the snapshot of the Colorado registration database. To avoid looking up the ORVR equipment
of individual make/model combinations in the transition model years of ORVR implementation,
we examined events for only the following combinations of model year and vehicle class: 2000-
2018 LDGVs, 2003-2018 LGDT12s, and 2006-2018 LDGT34s. Table 3-7 shows the model-year
distribution of Phase 1 plume observation results for these confirmed ORVR vehicles by vehicle
class.
If the Phase 1 plume observations of a refueling event did not indicate any plume in a
refueling event's videos, we did not need to re-view the videos for that event. As shown in Table
3-7, 1535 refueling events fell in this no-observed-plume category. Accordingly, we re-viewed
the videos of the refueling events in Table 3-7 only for light, heavy, and puddle Phase 1
observation results. The resulting dataset contained 455 refueling events made up of 1,373 30-
second videos.
3-26
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-13. A Frame with a Topping-Off Puff in a Video for the Rear Vehicle
Figure 3-14. A Frame with a Continuous Plume in a Video for the Front Vehicle
3-27
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-7. Phase 1 Status of Refueling Events for Confirmed ORVR Vehicles20'2122
LDGVs
Refu
eling E
vent P
ume
a)
Model
0)
c
Light
>
ra
~u
u
Total
Year
o
z
a)
X
3
a.
Events
2000
12
4
6
0
22
2001
10
4
6
0
20
2002
15
3
3
0
21
2003
17
6
2
0
25
2004
21
7
1
0
29
2005
27
11
5
0
43
2006
25
4
4
0
33
2007
37
9
2
0
48
2008
26
9
5
0
40
2009
35
6
3
0
44
2010
27
9
2
0
38
2011
27
5
1
0
33
2012
38
7
6
0
51
2013
50
10
0
0
60
2014
41
7
3
0
51
2015
35
4
1
0
40
2016
26
8
0
0
34
2017
28
5
1
0
34
2018
14
3
3
0
20
Total
511
121
54
0
686
LDGTIs + LDGT2S
Model
Year
Refu
(D
C
o
z
—
Light
m
vent P
>»
>
re
0)
X
ume
a)
~u
u
3
a.
Total
Events
2003
24
1
4
0
35
2004
33
6
3
0
42
2005
34
8
1
0
43
2006
39
10
3
1
53
2007
45
11
8
0
64
2008
39
6
3
0
48
2009
20
5
0
0
25
2010
31
9
0
1
41
2011
28
4
2
0
34
2012
47
6
2
0
55
2013
47
11
0
0
58
2014
51
3
5
0
59
2015
73
18
6
0
97
2016
48
14
3
0
65
2017
85
22
3
0
110
2018
64
5
2
0
71
Total
708
145
45
2
900
LDGT3S + LDGT4S
Model
Year
Refu
(D
C
o
z
—
Light =
m
vent P
>»
>
ra
a)
X
ume
a)
~u
u
3
a.
Total
Events
2006
17
3
2
0
22
2007
25
8
2
0
35
2008
18
4
2
0
24
2009
5
0
0
0
5
2010
14
2
3
0
19
2011
36
4
2
0
42
2012
29
5
2
0
36
2013
30
4
2
0
36
2014
38
6
4
0
48
2015
33
11
2
0
46
2016
22
6
3
1
32
2017
31
3
3
0
37
2018
18
3
1
0
22
Total
316
59
28
1
404
20 P:\EPA_RefuelingEmissions_WA2-23\Summer2019\Analysis\find_plumes_0RVR.sas
21 C:\Documents\EPA CanisterDegradation\WA2-23 (GasStnRebellion_MAR2019)\Report_Final/PlumeVideoCounts.xlsx
22 Vehicle classes have not been determined for 2019 vehicles in this table. In this table, the first model year in each vehicle class is the first full-
implementation model year for the class. Therefore, no non-ORVR vehicles and no vehicles in transition model years are counted for this table.
3-28
-------
High Evaporative Emissions Investigation Field Study
Final Report
We developed the Phase 2 evaluation method for re-viewing the videos of the selected
refueling events. Puffs of HC vapor are short-duration plumes that seem to be associated with the
customer's removal of the gas cap, beginning of gasoline flow, or end of fuel flow at nozzle
click-off After viewing all videos of the 455 selected refueling events, we discovered that
customers topping-off after the fuel nozzle had automatically clicked off also sometimes
produced puffs. Therefore, we went back again and re-viewed any refueling events where we had
seen continuous plumes during the first pass through during Phase 2. In several cases, some of
the initially assigned continuous plumes were actually multiple puffs caused by topping off.
Overall, the procedure was designed to characterize each of the six 5-second blocks
within each video. The Phase 2 instructions given in Appendix C were used to assign one of the
codes in Table 3-8 to each 5-second block in a video.
Table 3-8. Phase 2 Codes Used to Characterize Video 5-second Blocks
Code
Meaning
R
Puff at gas cap removal
B
Puff at beginning of fuel flow when the customer first pulls the nozzle handle
E
Puff at end of fuel flow when the nozzle clicks off
T
Puff caused by topping-off behavior after the nozzle automatically clicked-off
P
Puff coming from a puddle of gasoline on the pavement
1
Small, low-contrast, continuous plume
2
Small, high-contrast, continuous plume or a large, billowing, continuous plume
0
No plume and no puff can be seen
X
Screen is entirely white (from GCI calibration)
The codes for a 30-second video produce a 6-character string that summarizes what was
seen in the video. For example, 000R11 would indicate no emissions for about 15 seconds, a puff
at gas cap removal, followed by 10 seconds with a light plume. Obviously, this coding scheme
does not convey everything that can be observed in a video, but it does convey information about
refueling event emissions time trends for convenient analysis.
3-29
-------
High Evaporative Emissions Investigation Field Study
Final Report
While we recorded 6-character strings for each video, we also judged the Phase 2
category of the refueling event:
NoPuffsNoPlumes: We saw neither puffs nor plumes in any of the videos for the
refueling event,
OnlyPuffsNoPlumes: We saw at least one puff of any type (remove gas cap, begin fuel
flow, nozzle click-off, topping off, puddle), but we did not see any plumes associated
with periods of steady fuel flow, and
ContinuousPlumes: We did see plumes associated with periods of steady fuel flow, and
puffs of any type may or may not have been present.
OnlyPuffsNoPlumes as a category includes non-steady-state activities, such as, removing
a gas cap, beginning fuel flow, nozzle click-off, and topping off fuel tanks. These events are
likely included in the ORVR standard of 0.2 grams/gallon HC but would require further testing
to verify. We selected ContinuousPlumes as a category to measure the occurrence of refueling
events that today's evaporative emissions control systems are designed to control. Refueling
events in the ContinuousPlumes category might represent events that could be caused by
malfunctioning evaporative emissions control systems or canisters that are already partially
loaded.
3.6 Evaluation of GCI Camera Sensitivities in Phase 2 Plume Observations
The analysis in Section 3.2 revealed a bias in the model year trend of the Phase 1 fraction
of refuelings with observable emissions between the front and back pumps (see Figure 3-11 and
Figure 3-12). We attribute the difference in video-observable refueling emissions between front
and back pumps to a difference in the sensitivity of the GCI camera because of background
complexity and infrared lighting differences. Additionally, during that analysis, we hypothesized
that short-duration emission events might be more difficult to detect at front pumps than at back
pumps. Accordingly, to see if Phase 2 video viewings avoided, or at least reduced, the viewing
bias, we analyzed the Phase 2 viewing results for puffs, which tended to be short-duration events,
separate from continuous plumes, which tended to be longer duration events. For the analysis we
selected data for 2000-2018 LDGVs, 2003-2018 LDGT12s, and 2006-2018 LDGT34s, which all
have ORVR systems. This dataset contains 1,990 refueling events.
First, the analysis focuses on rates of Phase 2 continuous plumes at the front and back
pumps. Table 3-9 shows that the back-to-front ratio of continuous plume abundances was 1.6
(=4.8%/3.0%). That ratio is still larger than 1, but it is closer to 1 than the Phase 1 back-to-front
ratio of 2.4 (=30.2%/12.7%).
3-30
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-9. Continuous Plumes Observed at Front and Back Row Pumps23
Pump Row
Total Refuelings
(count)
Continuous
Plumes Observed
(count)
Continuous
Plumes Observed
(%)
Front
1037
31
3.0%
Back
953
46
4.8%
Total
1990
77
3.9%
Table 3-10 shows that the back-to-front ratios for R, B, and E puff types are 5.7
(=4.0%/0.7%), 3.4 (=12.4%/3.6%), and 2.3 (=14.2%/6.2%), respectively. All are substantially
greater than the Phase 2 continuous plume ratio of 1.6.
Table 3-10. Puff Types Observed at Front and Back Row Pumps
Pump
Row
Total
Refuelings
(count)
R: Gas Cap
Removal
Puff
(count)
B: Begin
Fuel Flow
Puff
(count)
E: Nozzle
Click-Off
Puff
(count)
R: Gas Cap
Removal
Puff
(%)
B: Begin
Fuel Flow
Puff
(%)
E: Nozzle
Click-Off
Puff
(%)
Front
1037
7
37
64
0.7%
3.6%
6.2%
Back
953
38
118
135
4.0%
12.4%
14.2%
Total
1990
45
155
199
2.3%>
7.8%
10.0%
Because more than one puff type can occur in a refueling event, we also consider the
abundance of one or more puffs. Table 3-11 shows that the back-to-front any-puff abundance
ratio is 2.5 (=24.2%/9.5%), which is still larger than the 1.6 ratio of the continuous plumes.
Table 3-11. Any Puffs Observed at Front and Back Row Pumps
Pump
Row
Total
Refuelings
(count)
Refuelings
with Puff
(count)
Refuelings
with Puff
(%)
Front
1037
99
9.5%
Back
953
231
24.2%
Total
1990
330
16.6%
Overall, the analysis results described above show that separating puff from plume results
diminishes the difference in the fraction of video-observable plumes of front versus back pumps.
Nevertheless, a difference between front and back pumps still exists. This does not mean that
results from front pumps are useless, invalid, or should be thrown out. It just means that GCI
camera videos of front pump refuelings tend to be less sensitive to imaging refueling emissions.
23 C:\Documents\EPA CanisterDegradation\WA2-23
(GasStnRebellion_MAR2019)\Analysis_Videos/ContinuousPlume Front v. Back.xlsx
3-31
-------
High Evaporative Emissions Investigation Field Study
Final Report
In Section 3.2, the analysis of Phase 1 video observations was used to evaluate the
influence of several factors on refueling emissions visibility in the GCI camera videos. Now, we
re-evaluate those factors for the plumes (not the puffs) identified by the Phase 2 re-viewing of
videos while using all of the Phase 2 confirmed ORVR data and while accounting for the
acknowledged difference in video sensitivity between front and back pumps.
We used logistic regression to determine the statistical significance of seven categorical
factors (wind speed, outdoor temperature, gallons of fuel dispensed, refueling time of day,
vehicle class, vehicle make, and distance between fuel pump and fuel fill door) after accounting
for front- vs. back-pump sensitivity. Figure 3-15 through Figure 3-21 show the distributions of
the factors for the confirmed ORVR dataset. Vehicle class, vehicle make, and distance between
fuel pump and fuel fill door are natural categorical variables as seen in Figure 3-19 through
Figure 3-21. However, the other four factors are continuous variables. To convert them to
categorical variables, we split each of the distributions at the 10, 25, 50, 75, and 90 percentiles
and created a high group and a low group for each split.
Table 3-12 shows the split point values for the four continuous factors. For example,
splitting the wind speed distribution using the 10-percentile value of 1.3 mph creates a high wind
speed group with observations with wind speeds greater than 1.3 mph and a low wind speed
group of observations with wind speeds less than or equal to 1.3 mph.
3-32
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-15. Under-Canopy Wind Speed during Confirmed ORVR Refuelings
Q.
E,
>>
CL
O
c
03
U
q3
"O
c
=5
¦o
CD
0)
Q.
CO
"D
C
0.2
0.6
1.0
1.4
1.8
2.2
2.6
3.0
3.4
3.8
4.2
4.6
5.0
5.4
5.8
6.2
6.6
7.0
7.4
7.8
8.2
8.6
9.0
9.4
9.8
10.2
10.6
11.0
11.4
11.8
12.2
12.6
13.0
CUM.
CUM.
Tl
73
m
FRE8>
¦38b
¦So
19
116
0.96
5.86
42
158
2.12
7.98
79
237
3.99
11.98
81
318
4.09
16.07
204
522
10.31
26.38
119
641
6.01
32.39
271
912
13.69
46.08
121
1033
6.11
52.20
155
1188
7.83
60.03
138
1326
6.97
67.00
99
1425
5.00
72.01
103
1528
5.20
77.21
67
1595
3.39
80.60
86
1681
4.35
84.94
48
1729
2.43
87.37
81
1810
4.09
91.46
39
1849
1.97
93.43
20
1869
1.01
94.44
19
1888
0.96
95.40
27
1915
1.36
96.77
25
1940
1.26
98.03
0
1940
0.00
98.03
4
1944
0.20
98.23
4
1948
0.20
98.43
9
1957
0.45
98.89
1
1958
0.05
98.94
7
1965
0.35
99.29
12
1977
0.61
99.90
0
1977
0.00
99.90
2
1979
0.10
100.00
0
1979
0.00
100.00
0
1979
0.00
100.00
0
100
200
300
FREQUENCY
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Arialysis/find_plumes_LR.sas 27FEB20 13:12
3-33
-------
High Evaporative Emissions Investigation Field Study Final Report
Figure 3-16. Under-Canopy Temperature during Confirmed ORVR Refuelings
a;
3
¦i—>
rp
a3
Q.
E
aj
o
o
"O
3
o
CUM.
CUM.
FREC^ FREq^
¦So
po9o
6
10
0.30
0.51
16
26
0.81
1.31
9
35
0.45
1.77
10
45
0.51
2.27
8
53
0.40
2.68
21
74
1.06
3.74
16
90
0.81
4.55
19
109
0.96
5.51
40
149
2.02
7.53
27
176
1.36
8.89
45
221
2.27
11.17
48
269
2.43
13.59
74
343
3.74
17.33
46
389
2.32
19.66
46
435
2.32
21.98
67
502
3.39
25.37
79
581
3.99
29.36
80
661
4.04
33.40
52
713
2.63
36.03
150
863
7.58
43.61
85
948
4.30
47.90
128
1076
6.47
54.37
70
1146
3.54
57.91
64
1210
3.23
61.14
80
1290
4.04
65.18
110
1400
5.56
70.74
96
1496
4.85
75.59
24
1520
1.21
76.81
103
1623
5.20
82.01
78
1701
3.94
85.95
115
1816
5.81
91.76
39
1855
1.97
93.73
65
1920
3.28
97.02
59
1979
2.98
100.00
160
FREQUENCY
/proj1/EPA_RefuclirigEmissions_WA2-23/Summcr2019/Analysis/1ind_plumes_LR.sas 27FEB20 13:12
3-34
-------
High Evaporative Emissions Investigation Field Study Final Report
Figure 3-17. Gallons Dispensed during Confirmed ORVR Refuelings
CUM.
10
c
_o
"ro
"O
0)
in
c
(1)
CL
LH
Q
d)
~o
t/1
03
C9
pcx
0.10
0.15
0.26
1.07
1.33
3.58
3.58
6.96
6.70
8.54
9.31
9.72
9.51
9.00
6.65
6.19
4.09
4.25
2.30
1.43
1.38
0.82
0.92
0.61
0.46
0.26
0.31
0.05
0.15
0.10
0.15
0.05
CUM.
PCX
0.10
0.26
0.51
1.59
2.92
6.50
10.08
17.03
23.73
32.28
41.59
51.30
60.82
69.82
76.47
82.66
86.75
91.00
93.30
94.73
96.11
96.93
97.85
98.47
98.93
99.18
99.49
99.54
99.69
99.80
99.95
100.00
20 40 60 80 100 120 140 160 180 200
FREQUENCY
/proj1/EPA_RefuclirigEmissions_WA2-23/Summcr2019/Analysis/1ind_plumes_LR.sas 27FEB20 13:12
3-35
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-18. Hour of the Day for Confirmed ORVR Refuelings
CO
~
CL>
J=
¦i—»
i-i—
o
s—
3
0
1
CUM.
CUM.
FREQ.
FREQ.
PCT.
PCT.
8
8
0.40
0.40
58
66
2.92
3.32
97
163
4.88
8.21
144
307
7.25
15.46
183
490
9.21
24.67
184
674
9.26
33.94
211
885
10.62
44.56
181
1066
9.11
53.68
158
1224
7.96
61.63
149
1373
7.50
69.13
145
1518
7.30
76.44
149
1667
7.50
83.94
135
1802
6.80
90.74
106
1908
5.34
96.07
78
1986
3.93
100.00
100 200 300
FREQUENCY
/proj 1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_LR.sas 27FEB20 13:12
3-36
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-19. Vehicle Class for Confirmed ORVR Refuelings
CUM. CUM.
FREQ. FREQ. PCT. PCT.
t/> LDGT12
i/>
_ro
U
a> LDGT34
.u
'jz
a)
> LDGV
0 100 200 300 400 500 600 700 800 900
FREQUENCY
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_LR.sas 27FEB20 13:12
3-37
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-20. Vehicle Make for Confirmed ORVR Refuelings
ACURA
AUDI
BMW
BUICK
CADILLAC
CHEVROLET
CHRYSLER
DODGE
FIAT
FORD
GMC
HONDA
HYUNDAI
n) INFINITI
JEEP
CO KIA
¦5 LAN DROVER
LEXUS
0) LINCOLN
u MAZDA
•p MERCEDES
m MERCURY
MINI
^ MITSUBISHI
NISSAN
OLDSMOBILE
PONTIAC
PORSCHE
RAM
SAAB
SATURN
SCION
SUBARU
SUZUKI
TOYOTA
VOLVO
VW
FREi
CUM.
FREO,
If
0.65
0.65
1.01
6.50
1.51
2.22
0.05
5.34
2.01
14.25
o18
2il1
oil
1.71
1.06
m
0.35
5.74
0.05
0.15
0.20
8:8s
0.40
0.10
6.55
0.10
25.78
0.86
1.06
CUM.
3.73
1T88
13.39
15.61
15.66
21.00
23.01
37.26
41.44
42.04
47.13
49.24
5§:55
53.93
55.64
56.70
56.85
57.70
63.85
64.00
64.20
65.11
65.16
65.56
6^6
72!31
98.09
98.94
100.00
100 200 300 400
FREQUENCY
500
600
/proj1/EPA_RefuclirigEmissions_WA2-23/Summcr2019/Analysis/1ind_plumes_LR.sas 27FEB20 13:12
3-38
-------
High Evaporative Emissions Investigation Field Study
Final Report
05
Figure 3-21. Pump to Fuel Fill Door Distance for ORVR Refuelings
ro
cu
0)
3
CUM. CUM.
FREQ. FREQ. PCT. PCT.
209 209 10.71 10.71
1742 1951 89.29 100.00
200 400 600 800 1000 1200 1400 1600 1800
FREQUENCY
Q.
E
3
Q_
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_LR.sas 27FEB20 13:12
3-39
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-12. Continuous Variable Split Values
Split
Point
(percentile)
Wind
Speed
(mph)
Outdoor
Temperature
(F)
Fuel
Volume
Dispensed
(gallons)
Refueling
Hour of
Day
(24-hour)
l<>
13
75
7.49
9
25
2.3
80
9.66
11
50
3.3
86
12.38
13
75
4.9
91
13.27
16
90
6.6
95
18.31
18
We used logistic regression to simultaneously determine the statistical significance of
pump position (front pump vs. back pump) and the factor category (for example, high wind
speed vs. low wind speed) on the probability that a plume (not a puff) would be observable in the
video. Thus, we performed twenty logistic regressions for the four variables and five splits
shown in Table 3-12. For the three natural categorical variables we performed an additional three
regressions.
In all 23 regressions, pump position (front vs. back) was statistically significant.
Specifically, we are 95% confident that the difference in average probabilities of continuous
plumes being seen in videos of refueling at the back pumps and at front pumps did not occur by
chance alone. Of the 23 regressions, after the effect of pump position was accounted for, only
one regression indicated a significant effect for the factor being investigated; all of the other
factors were found to have non-significant effects on the probability of an emissions plume being
seen in the video.
The one factor that showed an effect was the wind speed when 1.3 mph was used to split
the wind speed distribution into a low wind speed group and a high wind speed group. Table
3-13 shows the modeled observable plume probabilities for the four combinations of pump
position and wind speed when the wind speed distribution is split at 1.3 mph. The four cells in
the table also show the number of dataset observations for the plume and no-plume cases. For
example, for the low-wind speed, back-pump refueling condition, the table shows that the
logistic regression predicts that 8.0% of refuelings will produce a video with a visible plume.
That cell in the table shows that 7 plume videos and 75 no-plume videos were observed for that
refueling condition.
3-40
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-13. Logistic Regression Results for Wind Speed and Pump Position
Observable Plume Probability
Low Wind Speed
(< 1.3 mph)
High Wind Speed
(> 1.3 mph)
Back
Pumps
8.0%
Plume=7 NoPlume=75
4.3%
Plume=37 NoPlume=832
Front
Pumps
5.1%
Plume=6 NoPlume=122
2.7%
Plume=25 NoPlume=882
Analogous regressions when the wind speed distribution was split at higher speeds do not
show a significant effect on observable plume probability. Thus, it seems that the especially low
wind speeds (< 1.3 mph) particularly enhance the visibility of plumes in the GCI videos. In
summary, plumes are most visible in the GCI videos when a vehicle refuels at a back-row pump,
which tends to have a well illuminated and non-complex background, and when the wind is near
calm. Plumes are least visible at front-row pumps and when there is some air movement.
The quantified influences of pump position and wind on plume visibility make sense.
However, those factors cannot influence the refueling emissions themselves. The other six
factors (outdoor temperature, gallons of fuel dispensed, refueling time of day, vehicle class,
vehicle make, and distance between fuel pump and fuel fill door) could influence emissions, and
that is the reason they were explored with logistic regression. Higher outdoor temperatures and
more gallons dispensed would be more likely to result in saturated canisters. Refuelings early in
the day might be associated with smaller canister purge volumes if customers lived close to the
gas station. Refuelings near evening rush hour might be associated with higher fuel tank
temperatures and therefore higher fuel tank vapor generation. Some vehicle classes or vehicle
makes might be more likely to have inadequately designed evaporative emissions control
systems. When fuel pump hoses are stretched far to reach a fuel fill door on the opposite side of
the vehicle, refueling emissions could occur because of unusual orientations of the nozzle in the
fuel fill pipe. However, we saw none of these effects in the data since, as mentioned earlier, none
of the regressions on these factors were statistically significant.
3.7 Model Year Trends in Phase 2 Plume Observations
The Phase 2 re-examination of the videos for puffs and plumes effectively changes the
distribution of refueling emissions characteristics from the Phase 1 categories (Category 1,
Category 2, and Category 3) to the Phase 2 categories (NoPuffsNoPlumes, OnlyPuffsNoPlumes,
and ContinuousPlumes). To show and analyze the distribution shift, Table 3-14 shows the same
3-41
-------
High Evaporative Emissions Investigation Field Study
Final Report
counts of videos as were used to create the Phase 1 Table 3-7 but now using Phase 2 categories.
Table 3-14 again has the three major tables for LDGVs, LDGT12s, and LDGT34s, but the
subheadings refer to the three new refueling characteristics of Phase 2 viewing. In addition, we
show a NotAssigned category, which was created because a few of the refueling events could not
be unambiguously categorized. Categorization was not possible for these events because we
could not determine whether the emissions were from the target vehicle or from the other vehicle
also refueling on the same side of the island.
3-42
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-14. Phase 2 Status of Refueling Events for Confirmed ORVR Vehicles24 25
LDGVs
Refueling Event Category
Model
Year
NoPuffs
NoPlumes
Only Puffs
NoPlumes
Continuous
Plumes
Not
Assigned
Total
Events
2000
12
7
2
1
22
2001
11
3
5
1
20
2002
15
5
1
0
21
2003
19
4
1
1
25
2004
21
6
2
0
29
2005
29
9
5
0
43
2006
25
4
3
1
33
2007
41
5
2
0
48
2008
28
10
2
0
40
2009
37
4
3
0
44
2010
27
8
3
0
38
2011
27
4
2
0
33
2012
40
9
1
1
51
2013
52
7
1
0
60
2014
42
7
2
0
51
2015
36
3
1
0
40
2016
27
7
0
0
34
2017
30
4
0
0
34
2018
16
3
1
0
20
Total
535
109
37
5
686
LDGTIs + LDGT2S
Refueling Event Plume
Model
Year
NoPuffs
NoPlumes
Only Puffs
NoPlumes
Continuous
Plumes
Not
Assigned
Total
Events
2003
26
5
3
1
35
2004
35
6
1
0
42
2005
37
4
1
1
43
2006
41
7
5
0
53
2007
49
12
3
0
64
2008
40
4
3
1
48
2009
21
4
0
0
25
2010
31
9
1
0
41
2011
31
3
0
0
34
2012
47
7
1
0
55
2013
51
7
0
0
58
2014
51
7
1
0
59
2015
77
17
2
1
97
2016
50
12
3
0
65
2017
88
20
1
1
110
2018
67
4
0
0
71
Total
742
128
25
5
900
LDGT3S + LDGT4S
Refueling Event Plume
Model
Year
NoPuffs
NoPlumes
Only Puffs
NoPlumes
Continuous
Plumes
Not
Assigned
Total
Events
2006
17
3
2
0
22
2007
27
6
1
1
35
2008
19
4
1
0
24
2009
5
0
0
0
5
2010
14
4
1
0
19
2011
37
3
2
0
42
2012
32
3
1
0
36
2013
31
4
1
0
36
2014
38
8
1
1
48
2015
37
5
4
0
46
2016
24
7
1
0
32
2017
31
6
0
0
37
2018
20
2
0
0
22
Total
332
55
15
2
404
24 C:\Documents\EPA CanisterDegradation\WA2-23 (GasStnRebellion_MAR2019)\Report_Final/PlumeVideoCounts.xlsx
25 Vehicle classes have not been determined for 2019 vehicles in this table. In this table, the first model year in each vehicle class is the first full-
implementation model year for the class. Therefore, no non-ORVR vehicles and no vehicles in transition model years are counted for this table.
3-43
-------
High Evaporative Emissions Investigation Field Study
Final Report
Refueling Emissions Prevalence - We expect that results from only the
ContinuousPlumes category represent potential malfunctions of evaporative emissions control
systems.
Figure 3-22, Figure 3-24, and Figure 3-2626 show the model year trends for the fraction
of refueling events categorized as ContinuousPlumes. The other category that is of interest is the
OnlyPuffsNoPlumes category. Figure 3-23, Figure 3-25, and Figure 3-27 show the model year
trends for the fraction of refueling events categorized as OnlyPuffsNoPlumes for LDGV,
LDGT12s, and LDGT34s, respectively. Figure 3-22 through Figure 3-27 are made only for the
model years of those vehicle types where ORVR was required on all vehicles in the type. Figure
3-28 shows an overlay plot of the continuous plume model-year averages for the three vehicle
classes. Figure 3-29 shows an overlay of the puff trends seen in Figure 3-23, Figure 3-25, and
Figure 3-27.
The three ContinuousPlume plots show that each vehicle type tends to have a downward
trend as vehicles get newer. The downward trend is most obvious for the LDGVs in Figure 3-22
where 37 vehicles with ContinuousPlumes support the trend (see Table 3-14). The downward
trend in the Figure 3-26 plot for the LDGT34s is not so obvious because only 15 vehicles support
the trend. Since it may be that all three vehicle types have similar trends, we combined the 77
counts for all three vehicle types to create Figure 3-30. In the figure, the bubble symbols have
shading and areas proportional to the number of total refuelings for the model year. The data is
taken by combining the Table 3-14 ContinuousPlumes and Total Events data for the three
vehicle classes. For example, the symbol for 2002 is based on 21 refuelings (1 had a
ContinuousPlume) and 2015 is based on 183 refuelings (8 had a ContinuousPlume). The symbols
for 2006-2018 are for LDGVs, LDGT12s, and LDGT34s combined. The symbols for 2003-2005
are for only LDGVs and LDGT12s, and the symbols for 2000 and 2002 are for only LDGVs.
The symbols show a downward trend toward 0% ContinuousPlumes for new vehicles.
The three OnlyPuffsNoPlumes plots (Figure 3-23, Figure 3-25, and Figure 3-27) with the
open circles show that the model year trends for puffs caused by the combined effects of
removing gas cap, beginning fuel flow, click-off at the end of fuel flow, and topping off are
relatively flat with model year. This might be the expected trend since puffs are probably not
influenced by the evaporative emission control system. Also, note that the combined bubble plot
in Figure 3-31, using the OnlyPuffsNoPlumes and Total Events data from Table 3-14, shows that
26 C:\\Documents\EPA CanisterDegradation\WA2-23 (GasStnRebellion_MAR2019)\Report_Final/
PlumeVideoCounts.xlsx
3-44
-------
High Evaporative Emissions Investigation Field Study Final Report
the model-year average fraction of OnlyPuffsNoPlumes for all three vehicle types is about the
same at 14% (=243/1710). As for ContinuousPlume prevalence, the rate of puff occurrence
depends on camera sensitivity.
3-45
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-22. Model Year27 Trend of ContinuousPlume Fraction for LDGVs
0.14
O 0.12
o
5 0.10
O
re
t 0.20
to
§ 0.15
Q.
¦£0.10
0.05
0.00
p
o
5 C
o
o c
o
1
o
o
o
o
o
O
o
o
o
o
o
o
2000
2005 2010
Model Year
2015
2020
27
The data point for 2001 is off scale at 0.25.
3-46
-------
High Evaporative Emissions Investigation Field Study
Final Report
C
O
4->
o
ro
.
k
•—
• • T
•
—•—• 1 •
2005 2010
Model Year
2015
2020
Figure 3-25. Model Year Trend of OnlyPuffsNoPlumes Fraction
for 0-6,000 lb GVWR trucks (LDGT12s)
0.35
0.30
C
.2 0.25
+->
o
ra
£0.20
to
? 0.15
"I" 0.10
0.05
0.00
2000
T
o
o
c
) c o
O 0
o
° o o
1
3 o
o
o
1
2005 2010
Model Year
2015
2020
3-47
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-26. Model Year Trend of ContinuousPlume Fraction
for 6,001-8,500 lb. GVWR (LDGT34s)
0.14
I 0.12
'S
U
5 0.10
0 • •
£ 0.08
3
0.06
° #
= 0.04 •
1 #
o 0.02 •
o
0.00
2000 2005 2010 2015 2020
Model Year
Figure 3-27. Model Year Trend of OnlyPuffsNoPlumes Fraction
for 6,001-8,500 lb. GVWR (LDGT34s)
0.35
0.30
"c"
.2 0.25
o
re
£,0.20
to
§ 0.15
0_
0.10
o
0.05
0.00
2000 2005 2010 2015 2020
Model Year
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-28. Overlaid Model Year Trend of ContinuousPlume Fraction
0.14
-^0.12
O
ro 0.10
<1>
£ 0.08
3
Q_
•
r a
.
X
•
y.
if*x".
•
•
A L
A *
•
1 X A ** J
X
5
X
"•A
A A
A
A
• »x—1
• LGDVs
tf) 0.06 1 ** 1—• 1 ALDGT12S
3
O I A • ~ V I A • I X LDGT34S
J 0.04
'*->
c
o
O 0.02
2000 2005 2010 2015 2020
Model Year
Figure 3-29. Overlaid Model Year Trend of OnlyPuffsNoPlumes Fraction
0.35
0.30
| 0.25
g0.20
U) I ^ • rn » I V A ^ " I O LDGVs
A LDGT12S
Q_
= 0.10
o
0.05
>
O
o<
V
> *
a
o o
A
••a
O
^Aa
*
° AA
$
A
oa$a# o
O
<
>
<
O •
A
t LDGT34S
0.00
2000 2005 2010 2015 2020
Model Year
3-49
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-30. Model Year Trend of ContinuousPlume Fraction
for Combined LDGVs, LDGT12s, and LDGT34s
0.25
0.20
C
O
u
a>
E
0-
U)
3
O
D
c
c
o
u
0.15
0.10
0.05
0.00
'Bubble size and fill scale with count
)
-p_(
c
- 7 J-
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Model Year
/proj1/EPA_RcfiiolingEmissions_WA2-23VSummer2019/Analysis/find_plumcs_ORVR.sas 27FEB20 17:22
Figure 3-31. Model Year Trend of OnlyPuffsNoPlumes Fraction
for Combined LDGVs, LDGT12s, and LDGT34s
0.351
0.30
~ °-25
o
"J
u
£ 0.20
£ 0.15
>»
C
O
0.10
0.05
0.00
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Model Year
[
)
*Bubble size and fill scale with count
( )
I
J
M
r\ ( V—/
)
V J \ f \
J
/proj1/EPA_RcfualingEmissions_WA2-23VSummer2019/Analysis/find_plumcs_ORVR.sas 27FEB20 17:22
3-50
-------
High Evaporative Emissions Investigation Field Study
Final Report
We explored the continuous plume observation trends in the confirmed ORVR dataset
with ordinary least squares regression. For each of the 1,990 observations, the ContinuousPlume
variable had a value of either 1 (plume was observed) or 0 (plume was not observed). We
considered the influences of model year, vehicle type, and pump position (front vs. back).
Regressions indicated that model year had a statistically significant effect on the probability that
a refueling on an ORVR vehicle would produce an observable plume in a video. After model
year, regressions indicated a difference in the slope of the model-year trends between the front
pumps and back pumps. Finally, after model year and pump position were in the model, the
regressions found no significant differences in model year trends among the three vehicle types
(LDGVs, LDGT12s, LDGT34s). Therefore, we assumed that the model year trend was the same
for all three vehicle types and regressed all 1,990 observations on model year and pump position
together. The fitted trends produced by the regression are shown in Figure 3-32 with the 95%
confidence limits for the mean trend. The slope of the fit for the back pumps (red) is -0.56%/year
± 0.18%/year standard error and for the front pumps it is -0.41%/year ± 0.13%/year standard
error.
Figure 3-32. Regression of Continuous Plume Probability against Model Year
and Pump Position
Q.
in
3
o
c
o
U
n-
O
>
n
re
X!
o
0.08
0.02
0.00
2000
2005
2010
2015
2020
Model Year
Pump Position Front Pumps — 95% Confidence Interval — 95% Confidence Interval
Back Pumps — 95% Confidence Interval — 95% Confidence Interval
/proj1/EPA_RefuelingEmissions_WA2-23/Summer2019/Analysis/find_plumes_ORVR.sas 27FEB20 17:22
3-51
-------
High Evaporative Emissions Investigation Field Study
Final Report
Separate model year regressions for the back pumps and the front pumps also produced
the trend lines with the different slopes as seen in Figure 3-32. One feature of these trend lines is
that they intersect with each other and with the model-year axis around the 2019 model year.
This can be interpreted as indicating that the probability of observing continuous plumes on
brand new, 2019 vehicles is very low and possibly zero, and that whether the refueling occurs at
front pumps or at back pumps, that conclusion is the same.
For the analysis plots and figures in this section, we had not at first been able to decode
the VINs of the 2019 vehicles to get the vehicle classes. After the analysis was complete, we
were able to decode 56 of the 59 2019-model-year VINs used in Table 3-15. The results of the
Phase 2 video viewings of the 56 refuelings are shown in Table 3-15. No continuous plumes
were observed in any of the videos of the 2019 light-duty vehicle refuelings. Thus, this result is
consistent with the trend in Figure 3-30 and the notion that the probability of observing
continuous plumes in new vehicles is near zero. OnlyPuffsNoPlumes were seen in 18% (=10/56)
of the refuelings. This value is also consistent with the flat trend seen in Figure 3-31.
We suggest that the probability vs. model year trend lines pivot around the (2019, 0)
point as a function of the GCI camera viewing conditions. If the camera viewing conditions are
favorable to a high sensitivity by observing plumes, the trend line will be steep but will tend to
pass through, or near, (2019, 0). Conversely, if viewing conditions are not favorable, the trend
line will have a shallow slope but will still tend to pass through (2019, 0). The camera sensitivity
is affected not just by background illumination and complexity and wind speed; it is also affected
by the inherent sensitivity of the camera itself. Since we surmise that all gasoline vehicles
produce some, though perhaps tiny, refueling emissions, a sensitive camera would potentially
image plumes from all refuelings.
Thus, the important "take-away" from this analysis is that 1) brand new vehicles have a
near-zero probability of producing observable refueling plumes, and 2) as vehicles age, the
probability of having an observable plume increases approximately linearly. The slopes of the
model-year trends for the two sensitivities (front pumps and back pumps) shown in Figure 3-32
or for any particular camera or viewing condition sensitivity are not relevant since the slope
depends on the sensitivity of the camera and the viewing conditions. These considerations should
be taken into account when using this dataset for modeling.
3-52
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-
LDGVs
Model
Year
Refu
(/)
(/) 0)
H
n
eling Ev<
m
1 ^
Q- 3
£ °
o z
int Cate
0 f
1 E
~ 3
o
in
Not J
Assigned
Total
Events
2019
3
1
0
0
4
. Phase 2 Viewing Results for 2019 Light-
LDGTIs + LDGT2S
Model
Year
Ref
(/)
t/> 0)
4
n
c
OnlyPuffs =
NoPlumes to
m
<
Continuous 3
Plumes -g
ime
¦0
0)
c
O O)
Z w
!/>
<
Total
Events
2019
31
2
0
0
33
Vehicles
LDGT3S + LDGT4S
Model
Year
Ref
(/>
t/> O
4
n
c
OnlyPuffs =
NoPlumes in
m
vent Pli
(A
0 !T
1 E
¦¦p 3
§5:
0
ime
¦0
0)
c
O O)
Z w
(/)
<
Total
Events
2019
12
7
0
0
19
3-53
-------
High Evaporative Emissions Investigation Field Study
Final Report
Refueling Emissions Timing - The previous discussion reveals the prevalence of
refueling of ORVR vehicles with plumes and puffs. Now, we consider the timing of the plumes
and puffs. Since ORVR evaporative emissions control systems are required to control emissions
while refueling from 0 to 90% fuel tank levels but not greater than 90% full, ContinuousPlumes
that occur just before the nozzle automatic click-off may not represent control system
malfunctions. For a vehicle tank with a capacity of 20 gallons and a 10 gallon/minute fuel flow,
fueling from 90% to 100% full would occur for about the last 12 seconds before the nozzle
clicked off. Therefore, we might expect to see ContinuousPlumes in the last two 5-second blocks
before nozzle click-off
We determined the timing of plumes, puffs, and customer activities by viewing the
videos and recording the codes described in Table 3-8 for each 5-second block of the video. This
synthesized the video content into a short descriptor that contains the essential video information.
To help analyze the code strings for each video, we wanted to compare them to the Costco
transaction data that provided refueling event timestamps for credit card approval and fuel nozzle
hang-up. Potential refueling emissions were detected by the GCI camera collecting infrared data
and displaying it in a series of one to six 30-second Enhanced MidWave videos. The videos were
separated from each other by approximately 10-second gaps while storing data from the previous
video. Therefore, we concatenated the video code strings for the videos of each refueling event
and then compared the codes with the transaction timestamps to produce a combined time profile
of credit card approval and nozzle hang-up with the coded plume observations for each block.
Figure 3-33, Figure 3-34, and Figure 3-35 show time profile plots that represent the Phase
2 re-viewing results for refuelings of LDGVs, LDGT12s, and LDGT34s, respectively. The plots
are shown only for ORVR vehicles and only for those that we judged had at least one 5-second
block of ContinuousPlume no matter how weak or strong the emission plume appeared to be.
The plots are sorted by increasing model year within each vehicle type. The text beneath each
plot is a concatenation of the vehicle identifier, model year, make, model, empty vehicle weight,
vehicle class, and the 6-digit codes of the videoing viewings. The 6-digit codes should match the
solid lines and square symbols on the plot, as described below.
3-54
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-33. Refueling Emission Profiles with ContinuousPlumes
for LDGVs
8Q720T110597X9_2000_LEXUS_RX300_3900_LDGV_000000_000021_22ET00_
¦ L- Ar
10L2581E02V136 2000 LEXUS LX470 5300 LDGV 000000 2222ET 222200 000000
39G0182SW1S481_2001_INFINITI_QX4(SUV)_4300_LDGV_000000_000000_0XXX01_0XXXXX_000000
72K1460DC1 Y853_2001 _HONDA_CR-VEX 3300_LDGV_000000_0111XX_000000_
95YX274NC1I285 2001 TOYOTA Corolla/Matrix 2400 LDGV 000000 000011 2E0000
53Y7595PL0Z426 2001 NISSAN Sentra 2600 LDGV 000B00 11220E
0H160N17W00308_2001 _INFINITI_QX4(SUV)_4200_LDGV_000000_001212_222222_000000_
2O960B071782V2 2002 VOLVO V70 3900 LDGV 0B0000 000000 000000 000000 2E1000
Jx/h
5F830Q27E190B4_2003_MITSUBISHI_GalantES/GTZ/LS_3100_LDGV_010121_000000_
-A-
2P966W081968A83 2004 VOLVO XC90 4700 LDGV 000200
PLOT ¦ ¦ ¦ GasCapRemoval_Puff
¦ ¦ ¦ Puddle_Puff
± ^ ~ Nozzle_HangUp
Begin Fuel Flow_Puff
ToppingOff_Puff
Continuous Plume
¦ ¦ EndFuelFlow_Puff
A A CreditCard_Approval
3-55
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-33 (continued). Refueling Emission Profiles with ContinuousPlumes
for LDGVs
; . va —\—a
18T9430V09Y532_2004_TOYOTA_Prlus_2900_LDGV_0B2120_000000
f
75V1685NL2L128 2005 NISSAN Sentra 2600 LDGV 000000 000002 E00000
59Z3286QF7Y780_2005_MERCEDES_C230_3200_LDGV_RR1100_000000_11ETT1_
07B5310F11L286_2005_AUDI_TT_3200_LDGV_000000_000000_001222_200000_
30Q5750V27Y982_2005_VOLVO_S40/V50_3500_LDGV_000222_222200_
6TCX2W7OF6H4C2_2006_MERCEDES_C280_3600_LDGV_0B0000_111010_000000_000000_000000
43T1732E70N700_2006_SUBARU_Legacy/Outback_3400_LDGV_000000_000010_221E00_
7Q0X0X18G515J5_2006_FORD_500SELFWD_3600_LDGV_0B0000_000000_1122E0_
52X9905HU3V071_2007_TOYOTA_Camry_3400_LDGV_000000_000001_000000_
Ju
60729D92F111B2 2007 MERCEDES C280 3600 LDGV 000000 000000 01TTT0
PLOT ¦ ¦ ¦ GasCapRemoval_Puff
¦ ¦ ¦ Puddle_Puff
aaa Nozzle_HangUp
Begin Fuel Flow_Puff ¦ ¦ ¦ EndFuelFlow_Puff
Top ping Off_Puff a a a CreditCard_Approval
Continuous Plume
3-56
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-33 (continued). Refueling Emission Profiles with ContinuousPlumes
for LDGVs
62CX380UC9P779 2007 HONDA CR-VEX4WD 3500 LDGV 0000B0 000000 000012 E00000
17V519XTC0Y138 2008 BMW 535xi 4200 LDGV 000000 000000 222E00
4E210C132536C3_2008_LEXUS_ES350_3600_LDGV_000000_000000_2222E0_
02PX100YZ3R652 2009 TOYOTA Corolla 2800 LDGV 100000 000000
AJ L
4l633Q12A441M1_2009_HONDA_AccordLX_3200_LDGV_011110_010TT0
4l633Q12A441M1_2009_HONDA_AccordLX_3200_LDGV_000000_110001_111E00_
_r
9F702Q16U060J7_2010_CADILLAC_DTS_4000_LDGV_000000_12E000
9D763T06C472O7 2010 ACURA TSX 3400 LDGV 022111 000000
28 J1405XC3W122_2010_TOYOTA_Corolla_2700_LDGV_000000_222222_2E1000_
y.
40QX776IH1Y088_2011_DODGE_ChargerR/T_4400_LDGV_000000_000000_012ETT
PLOT ¦ ¦ ¦ GasCapRemoval_Puff
¦ ¦ ¦ Puddle_Puff
a a a Nozzle_HangUp
Begin Fuel Flow_Puff
Top ping Off_Puff
Continuous Plume
¦ ¦ EndFuelFlow_Puff
A A CreditCard_Approval
3-57
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-33 (continued). Refueling Emission Profiles with ContinuousPlumes
for LDGVs
73PX473TM5Q120_2011_INFINITI_G25/G37Coupe_3700_LDGV_000000_000100_000000_111100_
~*r
~l 1 r
94TX332RH4Y434_2012_SUBARU_lmpreza_3100_LDGV_000000_000000_222E00_
3L419Q54N111K7_2013_NISSAN_Altima_3100_LDGV_000010_000000_11111E_
5O330D53D328B1 2014 BMW 535ixDrive 4300 LDGV 010000 000000 E00000
6207614QC8E038_2014_NISSAN_Maxima_3500_LDGV_000000_000000_1222E0_
98Z0222JM0Q237_2015_FORD_FiestaST_2720_LDGV_lXXXXX_000000_
4N349B18A712Y5_2018_HC)NDA_AccordSport_3200_LDGV_100000_00000E
PLOT ¦ ¦ ¦ GasCapRemoval_Puff ¦ ¦ ¦ Begin Fuel Flow_Puff ¦ ¦ ¦ EndFuelFlow_Puff
¦ ¦ ¦ Puddle_Puff ¦ ¦ ¦ ToppingOff_Puff a a a CreditCard_Approval
* ^ ± Nozzle_HangUp Continuous_Plume
3-58
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-34. Refueling Emission Profiles with ContinuousPlumes
for LDGT12S
01N909BHK7Y595_2003_FC)RD_EscapeXLT_3400_LDGT12_000000_000000_000000_000000_210000
6L052Q12Z513G3_2003_TOYC>TA_TacomaDeluxe_3600_LDGT12_000000_222222
1
n r
1603071F01 J592_2003_TOYOTA_Highlander_3600_LDGT12_001111_222220_000000
5O734D24T720E5 2004 CHRYSLER PTCruiserGT/Dream 3300 LDGT12 000000 000000 001011 T00000
2F448C24S86604 2005 TOYOTA SlennaLE 4200 LDGT12 000000 122222
4U502N51H12706_2006_HONDA_Pilot_4500_LDGT12_000000_000000_000110_
jU
57W1765JC0Z501 2006 NISSAN Xterra 4300 LDGT12 000000 000000 000012
41D6791LW7W424_2006_JEEP_LibertySport4WD_3900_LDGT12_000000_000000_000000_011122_
76V1294KC3A189 2006 NISSAN Frontier 4300 LDGT12 122210
14H2985S56X012 2006 KIA Sorento4WD 4400 LDGT12 001222 220000
PLOT ¦ ¦ ¦ GasCapRemoval_Puff
¦ ¦ ¦ Puddle_Puff
± ^ ~ Nozzle_HangUp
Begin Fuel Flow_Puff
ToppingOff_Puff
Continuous Plume
¦ ¦ EndFuelFlow_Puff
A A CreditCard_Approval
3-59
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-34 (continued). Refueling Emission Profiles with ContinuousPlumes
for LDGT12S
1 U. 1 I ^
50828Q07B416U2_2007_HONDA_Pilot_4400_LDGT12_110000.
84J7776YC1P369_2007_NISSAN_Pathfinder_4600_LDGT12_000000_000001_210000_000000_
66X7210Z51N848_2007_TOYOTA_RAV4Sport_3600_LDGT12_000000_2222E0
3Q2X5M11W617Q5_2008_JEEP_Libertyl_imitedEdi_4500_LDGT12_000XXX_000000_000000_000221_
61G7955GZ5S888 2008 TOYOTA TacomaDLX 4100 LDGT12 000000 000011 22222E 100000
, n
i
3009071104R237 2008 MAZDA CX-9 4600 LDGT12 000000 002220 222222 E20000
42M4504YC20552 2010 NISSAN Frontier 4500 LDGT12 000012
w
3Q719Q07L342B4_2012_HONDA_CR-V_3500_LDGT12_000000_222200_
5Q129Z203730M2_2014_SUBARU_Outback_3600_LDGT12_000000_001222_E00000
73Y422BGB2G743_2015_FORD_Edge_4100_LDGT12_000000_000001_1E0000
PLOT ¦ ¦ ¦ GasCapRemoval_Puff
¦ ¦ ¦ Puddle_Puff
* ^ a Nozzle_Hangllp
Begin Fuel Flow_Puff
ToppingOff_Puff
Continuous Plume
i ¦ ¦ EndFuelFlow_Puff
A A CreditCard_Approval
3-60
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-34 (continued). Refueling Emission Profiles with ContinuousPlumes
for LDGT12S
60345C69H836E3 2015 HONDA CR-V 3500 LDGT12 000000 000001 222120
05CX427ZM2X775 2016 HONDA HR-V 3100 LDGT12 RRRRRR 212000 000000
06C6848R72R618_2016_KIA_Sorento/Sportage_3600_LDGT12_000001_111010_000000_
0Y013P08D468Q6_2016_TOYOTA_RAV4Hybrid_4000_LDGT12_000122_000000
1O994MB5B698E7_2017_FORD_Edge_4100_LDGT12_21E000_
PLOT ¦ ¦ ¦ GasCapRemoval_Puff ¦ ¦ ¦ Begin Fuel Flow_Puff ¦ ¦ ¦ EndFuelFlow_Puff
¦ ¦ ¦ Puddle_Puff ¦ ¦ ¦ ToppingOff_Puff a a a CreditCard_Approval
aaa Nozzle_HangUp Continuous_Plume
3-61
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-35. Refueling Emission Profiles with ContinuousPlumes
for LDGT34S
™—i 1 r
020X79 JCF0E290_2006_LINCOLN_MarkLT4WDSuperCre_5500_LDGT34_000000_000000_11110X
V
38L7815NS0Q248_2006_TOYOTA_Tundral_imited_4900_LDGT34_222222_222222_222200_
1A168M62Z354Q5_2007_CHEVROLET_15004WD_5300_LDGT34_000000_000000_000001_000000_0022EE
j- _r~Ln_
2K223C19L666A8 2008 BMW_X54.8i 5600_LDGT34 011111 000001 011010 000000
36E8476QS9T758_2011_DC)DGE_RAMPickupLightDut_5200_LDGT34_00B000_000000_000000_000000_010000
6J828D38R212O7_2011_CHEVROLET_SuburbanLT_5700_LDGT34_00B000_001112_0EE000_
8Q026Z11B32804_2012_HONDA_Odyssey_3700_LDGT34_000000_000000_000000_212222_
I
3M557GC9G623Q6_2013_Ford_UtilityPolicelnte_._LDGT34_000000_000222_22222E_200000_
71 D0350QB2N502_2014_HONDA_Pilot_4500_LDGT34_000000_000111_000000_
95Q8055HR1S990_2015_GMC_YukonDenali_5800_LDGT34_000000_000000_000100_000000_
PLOT ¦ ¦ ¦ GasCapRemoval_Puff
¦ ¦ ¦ Puddle_Puff
a a a Nozzle_Hangllp
Begin Fuel Flow_Puff
ToppingOff_Puff
Continuous Plume
¦ ¦ EndFuelFlow_Puff
a a CreditCard_Approval
3-62
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 3-35 (continued). Refueling Emission Profiles with ContinuousPlumes
for LDGT34S
1 1 r
6RO30W1DR4X2D7G_2015_CHEVROLET_SuburbanLT_5900_LDGT34_000B00_000000_000001_000000_
IT
65L1725TR4Z486 2015 CHEVROLET SuburbanLS 5800 LDGT34 000000 111101 0EE000
_rt
_n_
30A1531 AC3C394_2015_DODGE_Durango_4900_LDGT34_000000_001120_000010_1122ET_
00787023R144C5_2016_CHRYSLER_Town&Country_4500_LDGT34_000000_222E21_
PLOT ¦ ¦ ¦ GasCapRemoval_Puff
¦ ¦ ¦ Puddle_Puff
a a a Nozzle_Hangllp
Begin Fuel Flow_Puff
ToppingOff_Puff
Continuous Plume
¦ ¦ End Fuel Flow_Puff
A A CreditCard_Approval
3-63
-------
High Evaporative Emissions Investigation Field Study
Final Report
Since these plots are unconventional, they require some explanation. The x-axis is time
since the vehicle arrived at the pump. The gray vertical grid lines are spaced 30 seconds apart.
The y-axis is the scale for ContinuousPlume contrast, which is plotted using the solid black line.
The bottom y-axis tick mark represents 0 (no plume), and the top tick mark represents 2 (high
contrast or large billowing plume). The trend of the plotted black line reveals the trend of the
ContinuousPlume. Each black line segment is 30 seconds long with gaps of about 10 seconds
between the segments. This reflects the 30-second videos and 10-second gaps between videos.
No emissions information can be obtained during the gaps since there is no video to observe
during that time.
The symbols on the plots represent Costco transaction events (triangles) and puffs
(squares). The blue triangle marks the time when the customer's credit card was approved by the
gas pump. The start of fuel flow must occur after the approval, but it may not occur immediately
after approval. The black triangle marks the time when the fuel nozzle is hung up on the fuel
pump at the end of refueling. Again, the time between the end of fuel flow and the hang-up may
be short or long, but the end of fuel flow must be before the hang-up.
The square symbols represent puffs that we observed in the videos at gas cap removal
(black), at the beginning of fuel flow (green), at the nozzle click-off (red), at customer topping-
off activities (purple), and from a gasoline puddle on the pavement (orange, note that none were
seen in the continuous plume samples plotted in these figures, but the symbol is included for
completions since they were present in the puffs seen). Keep in mind that these symbols mark
those activities only if we saw puffs during those activities. So, if we did not see a puff or if we
did not see or could not determine when the activity occurred, there is no symbol for it.
Vehicles with Repeat Refuelings - Since the study extended over 3 weeks, there was a
chance that we might obtain videos on vehicles that returned to the Arvada station for repeat
refuelings. We found that of the 1,990 ORVR light-duty vehicle refuelings, 111 refuelings were
repeat refuelings of 55 vehicles. 54 vehicles came twice. Of these 54 vehicles, 37 had
NoPuffsNoPlumes both times. The other 17 of the 54 vehicles had NoPuffsNoPlumes one time
and OnlyPuffsNoPlumes the other time. The one vehicle that came three times had
ContinuousPlumes two times and NoPuffsNoPlumes the third time. The time profiles for the
ContinuousPlume events for this vehicle (a 2009 Honda Accord) are shown in the fifth and sixth
plots on the third page of Figure 3-33.
To demonstrate the variability of refueling emissions and the variety of conditions that
make explaining plume visibility difficult, we examine the details of the 2009 Honda Accord in
3-64
-------
High Evaporative Emissions Investigation Field Study
Final Report
more detail. This vehicle actually came four times (July 10, 17, 20 and 21) to the fuel island that
we were monitoring; it is possible that it came on other occasions to fuel islands that we were not
monitoring. On July 17 the refueling event was not videoed, so we cannot determine refueling
emissions. On July 10, 20, and 21, this was the respective information: ContinuousPlume,
NoPuffsNoPlumes, ContinuousPlume; 10.4, 8.9, 9.0 gallons dispensed; front, back, front pump;
3.3, 5.1, 0.2 mph wind; NE, N, S wind direction.
Refueling Mass Emissions Rate - The original plans for this study called for
quantification of the mass of HC present in the Rebellion Photonics infrared data. This data
analysis activity was not carried out.
The GCI camera records a large amount of "raw" infrared spectral data for all of the
camera's pixels each 1/15 second. The 30-second Enhanced MidWave videos that we have used
for this analysis were produced on site from this raw data. The plan called for Rebellion to post-
process the raw data to produce ColoredVIS videos for 2,000 selected 30-second segments.
ColoredVIS videos are 15 frame/second videos made up of conventional visible-range video
overlaid with a false coloration of the plume. ColoredVIS videos are routinely made by
Rebellion for their other clients. The coloration is based on the optical mass (ppm-m) measured
by the GCI camera. Thus, a ColoredVIS video would present all of the optical information that
the camera detected in the context of the HC emissions. The processing could also output the
optical mass for each pixel and each video frame for those pixels that the post-processing
determined were part of the emissions plume. We would not have been able to determine the
refueling emission rate, but we would have been able to determine the mass of HC visible to the
camera at any given instant. The plan called for first checking the efficacy of the post-processing
to make ColoredVIS videos by applying the technique to the GCI camera data collected on the
reference vehicle runs since we knew the emissions concentrations and flow rates for those runs.
We submitted 18 reference vehicle runs to Rebellion to test the capability of Col or VIS
video production. Unfortunately, we judged that ColorVIS post-processing would not be able to
produce useful information for the study. We found that while the post-processing could detect
strong HC plumes, it was poor at detecting weak ones. Additionally, the processing always
falsely assigned large artifacts to the plume thus hugely elevating the plume optical mass. This
unacceptable behavior was caused by the complex background in the gas stations scenes: moving
vehicles, moving people, wisps of plumes, plumes obscured by opening car doors, and
inconsistent background lighting (sometimes brightly lit, sometimes in the shade of the canopy).
Accordingly, we decided that the possible benefit of plume quantification could not be
reasonably achieved within the desired accuracy and within the budget.
3-65
-------
High Evaporative Emissions Investigation Field Study
Final Report
3.8 Investigation of Refueling Plumes from Medium-Duty Vehicles
During analysis of the Arvada gas station data, EPA asked us if there were any refuelings
on medium-duty vehicles. Since we had removed motorhomes and buses from the Colorado
vehicle registration snapshot look-up table, medium-duty vehicles did not appear in the study's
master dataset. To answer the question, we wrote a special SAS program28 to search for medium-
duty vehicles. After filtering, we found five motorhomes and a gasoline "bus" that had not been
in the ExE listing before. When these six vehicles were added to the others that were already in
the ExE listing, there were a total of 59 vehicle refuelings with vehicle classes HDGV2b, HDGB,
HDGV3, and HDGV4, as shown in Table 3-16,29 Table 3-17, Table 3-18, and Table 3-19,
respectively.
Refueling data was captured on 43 HDGV2b vehicles, as shown in Table 3-16. These
refuelings all appear to be on complete vehicles (not incompletes). For this class: pre-ORVRs
were MY 2003 and before, implementation transition years were MY 2004-2005, and ORVR
were MY 2006 and after. The table shows 13 pre-ORVR refuelings with 10 (77%) having Light
or Heavy Phase 1 emissions visible in the videos. For this vehicle class, there were 27 ORVR
refuelings with 8 (30%) refuelings with Light or Heavy Phase 1 emissions visible.
Four of the 27 refuelings were by one individual vehicle on different days: JUL 15, 16,
22, and 23. The first two refuelings had ContinuousPlumes, the third had OnlyPuffsNoPlumes,
and the fourth had NoPuffsNoPlumes. The volumes of fuel dispensed were quite consistent:
16.1, 15.9, 16.9, and 16.6 gallons. The pump positions for the four refuelings were front, front,
back, and front, respectively. The details of the repeated refuelings do not provide clarity for the
reason that plumes and puffs were sometimes seen and sometimes not.
Refueling data was captured on 5 HDGV3 vehicles, as shown in Table 3-17. Three of
these refuelings appear to be on complete vehicles, and two are on incompletes, which happened
to be recreational vehicles. For this HDGV3 class, ORVR is not required until the 2017 model
year; however, manufacturers are believed to commonly install ORVR systems according to the
HDGV2b schedule - unless they are incompletes. Therefore, we expect that the two incompletes
have no ORVR systems. We are uncertain whether the three completes in the table have ORVR
systems of not. Taken altogether, the 4 out of 5 HDGV3 vehicles produced Heavy Phase 1
emissions.
28 P:/CDPHE/Regis2019/REGmissing.sas
29 C:\Documents\EPA CanisterDegradation\WA2-23
(GasStnRebellion_MAR2019)\Analysis_Videos/MDV_masterlist-200217.xlsx
3-66
-------
High Evaporative Emissions Investigation Field Study
Final Report
Refueling data was captured on 4 HDGV4 vehicles, as shown in Table 3-18. All of these
vehicles are incompletes and are before the 2017 model year when HDGV4 completes must have
ORVR systems. Therefore, we expect that all four vehicles probably do not have ORVR
systems. The Phase 1 results in the table show that all four refueling produced heavy emissions.
The refuelings for HDGB vehicles are shown in Table 3-19. Three of the vehicles are
before 1999 and therefore are likely pre-ORVR. All 3 of these refuelings produced Phase 1 Light
or Heavy emissions in the videos. The other 4 refuelings are for HDGVs with model years 2012
and newer. Only one of these produced a Light Phase 1 emission.
We considered the Phase 2 re-viewings only on refuelings when ORVR systems are
expected to be on vehicles. Accordingly, we did not re-view videos for Phase 2 results if we
knew that the vehicle was pre-ORVR or no-ORVR, which is the reason that some cells in the
Phase 2 results in the tables are blank.
For the HDGV2b ORVR vehicles, Table 3-16 shows that 3 (11%) of 27 refuelings had
ContinuousPlumes. Two of those continuous plumes were produced by one vehicle. Adjusting
for this and stating the results by vehicle, 2 (8%) of 24 vehicles has ContinuousPlumes.
Table 3-17 for the HDGBs shows that none of the 4 non-pre-ORVR HDGB vehicles
produced ContinuousPlumes.
In Table 3-18, both of the two newest HDGV3 vehicles had ContinuousPlumes, but we
could not assign ORVR equipment status to these vehicles.
Since all of the four HDGV4 vehicles in Table 3-19 were incompletes, they are not
expected to have ORVR systems. The three vehicles with Phase 2 results indicated
ContinuousPlumes.
3-67
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-16. Refueling Plume Results for HDGV2b
Vehicle
Class
ORVR
Assignment
Vehicle Description
Phase 1
Emissions
Summary
Phase 2
Emissions
Summary
VehID
i_Depart_MTN
(LinkTime)
HDGV2b
pre-ORVR
1990 FORD F250RegularCab 4400
H
10Y619ATC6Y811
18JUL19:20:12:37
HDGV2b
pre-ORVR
1994 FORD F250SuperCab4WD 5700
H
0F055PB3K064O9
18JUL19:14:25:53
HDGV2b
pre-ORVR
1999 FORD F250SuperCab4WD 5600
H
91Z341AEE9N628
23JUL19:14:45:34
HDGV2b
pre-ORVR
1999 FORD F250SuperCab4WD 5700
L
0E620CD5E380O3
22JUL19:17:18:39
HDGV2b
pre-ORVR
2000 CHEVROLET Express35002WD 5300
H
9L555O201185E5
10JUL19:08:44:00
HDGV2b
pre-ORVR
2000 FORD ExcursionLimited4 6900
H
0U5X6MB7E127O2
22JUL19:16:41:04
HDGV2b
pre-ORVR
2001 CHEVROLET K2500Pickup4WD 6700
0
NoPuffsNoPlumes
05Z3211LF0X084
12JUL19:18:13:15
HDGV2b
pre-ORVR
2001 FORD E2502WD 5100
H
75V007BQH3C694
23JUL19:12:09:44
HDGV2b
pre-ORVR
2001 FORD F2502WD 4900
0
NoPuffsNoPlumes
21W120BQE8Z401
15JUL19:13:12:27
HDGV2b
pre-ORVR
2002 CHEVROLET 3500Van2WD 6300
H
5H776R171131Q7
18JUL19:14:53:41
HDGV2b
pre-ORVR
2002 DODGE RamVan/Wagon3500 4800
H
0H548R11K49102
08JUL19:09:56:28
HDGV2b
pre-ORVR
2002 GMC Sierra2500Pickup4 6200
0
NoPuffsNoPlumes
98F4852VE9I454
22JUL19:16:13:27
HDGV2b
pre-ORVR
2003 FORD F250SuperDuty4WD 5800
H
73Y383ALE7X290
16JUL19:19:20:40
HDGV2b
transition
2004 FORD E2502WD 5100
0
NoPuffsNoPlumes
00112KB1H268G1
09JUL19:08:11:28
HDGV2b
transition
2005 CHEVROLET K2500Pickup4WD 6000
H
7F097Q81F266B1
09JUL19:11:59:18
HDGV2b
transition
2005 GMC Sierra2500Pickup4 5700
H
38H0352UE3Y486
21JUL19:10:06:28
HDGV2b
ORVR
2007 CHEVROLET 25004WD 5800
0
NoPuffsNoPlumes
8H265Q16E928I8
09JUL19:15:01:11
HDGV2b
ORVR
2007 CHEVROLET 25004WD 5900
0
NoPuffsNoPlumes
76S4515XE1S241
09JUL19:17:38:46
HDGV2b
ORVR
2007 FORD E2502WD 5200
0
NoPuffsNoPlumes
3I093ZB2D386Q4
12JUL19:12:23:43
HDGV2b
ORVR
2008 FORD E2502WD 5206
0
NoPuffsNoPlumes
6B145DA8D232F4
15JUL19:18:50:53
HDGV2b
ORVR
2008 FORD F2504WDSRW 6600
L
OnlyPuffsNoPlumes
50194ND2E75 8T9
12JUL19:10:21:57
HDGV2b
ORVR
2008 FORD F2504WDSRW 6600
0
NoPuffsNoPlumes
89H029DKE9U569
23JUL19:19:20:23
HDGV2b
ORVR
2008 FORD F250SuperDuty4WD 6300
0
NoPuffsNoPlumes
7Q397LA7E098Z9
12JUL19:12:57:48
HDGV2b
ORVR
2009 CHEVROLET Silverado/Suburba 5700
0
NoPuffsNoPlumes
86X0531NE2O507
21JUL19:11:40:51
HDGV2b
ORVR
2009 FORD F250SupercabSRW4W 6300
H
ContinuousPlume
6E061CA8E444H6
21JUL19:08:54:56
3-68
-------
High Evaporative Emissions Investigation Field Study
Final Report
Vehicle
Class
ORVR
Assignment
Vehicle Description
Phase 1
Emissions
Summary
Phase 2
Emissions
Summary
VehID
i_Depart_MTN
(LinkTime)
HDGV2b
ORVR
2010 CHEVROLET Silverado2500 5800
L
OnlyPuffsNoPlumes
9709541TZ7Z205
18JUL19:12:20:59
HDGV2b
ORVR
2010 FORD EconolineE350 5400
0
NoPuffsNoPlumes
1Q286ZA9D591K7
18JUL19:18:07:47
HDGV2b
ORVR
2010 FORD EconolineE350 9500
0
NoPuffsNoPlumes
04Y599AJD6H683
23JUL19:16:24:49
HDGV2b
ORVR
2011 FORD EconolineE350 5200
0
NoPuffsNoPlumes
4Q578KA7D364I3
22JUL19:07:24:34
HDGV2b
ORVR
2014 NISSAN NV1500/NV2500/NV3 6100
0
NoPuffsNoPlumes
38A0711IN0V413
16JUL19:08:50:03
HDGV2b
ORVR
2015 FORD F250 6500
0
NoPuffsNoPlumes
72V845CQE1R228
10JUL19:09:13:36
HDGV2b
ORVR
2015 FORD Transitu50 4800
L
NoPuffsNoPlumes
62L236BGK2Q276
09JUL19:09:31:05
HDGV2b
ORVR
2015 FORD TransitT250 5000
0
NoPuffsNoPlumes
80Q569ASK7T900
15JUL19:17:56:43
HDGV2b
ORVR
2015 GMC Savana3500 5700
L
OnlyPuffsNoPlumes
80P2732X13S620
23JUL19:12:06:38
HDGV2b
ORVR
2016 FORD TransitT250 4800
0
NoPuffsNoPlumes
8O203CA8K293R1
16JUL19:11:50:48
HDGV2b
ORVR
2016 NISSAN NV1500/NV2500/NV3 6000
H
ContinuousPlume
0U009J89N123Q8
15JUL19:10:23:36
HDGV2b
ORVR
2016 NISSAN NV1500/NV2500/NV3 6000
H
ContinuousPlume
0U009J89N123Q8
16JUL19:11:18:47
HDGV2b
ORVR
2016 NISSAN NV1500/NV2500/NV3 6000
H
OnlyPuffsNoPlumes
0U009J89N123Q8
22JUL19:18:33:04
HDGV2b
ORVR
2016 NISSAN NV1500/NV2500/NV3 6000
0
NoPuffsNoPlumes
0U009J89N123Q8
23JUL19:11:04:58
HDGV2b
ORVR
2017 GMC Sierra2500 6700
0
NoPuffsNoPlumes
1N503B21F428W3
10JUL19:17:42:36
HDGV2b
ORVR
2017 NISSAN NV1500/NV2500/NV3 6200
0
NoPuffsNoPlumes
5I864A83N068O8
15JUL19:17:10:08
HDGV2b
ORVR
2018 FORD TransitT150 3968
0
NoPuffsNoPlumes
4D886CA9K230Z4
18JUL19:18:14:54
HDGV2b
ORVR
2018 FORD TransitT250 4934
0
NoPuffsNoPlumes
3E995CB6K550L7
16JUL19:10:24:49
3-69
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-17. Refueling Plume Results for HDGBs
Phase 1
Phase 2
Vehicle
ORVR
Emissions
Emissions
i_Depart_MTN
Class
Assignment
Vehicle Description
Summary
Summary
VehID
(LinkTime)
HDGB
pre-ORVR
1994 FORD E350SuperWagon 6100
H
1D9X0CA5H415N9
09JUL19:09:59:12
HDGB
pre-ORVR
1996 FORD E350SuperWagon 5900
L
71H663BUH5C837
12JUL19:15:25:10
HDGB
pre-ORVR
1998 FORD E350SuperWagon 6100
H
6P168WA8H289B54
22JUL19:12:02:57
HDGB
?
2013 FORD EconolineE350 5800
0
NoPuffsNoPlumes
0 V6460A4D5 79H0
23JUL19:14:51:49
HDGB
?
2014 FORD EconolineE350 5800
L
OnlyPuffsNoPlumes
51S066AQD0W454
12JUL19:20:26:16
HDGB
?
2015 FORD TransitT350 5900
0
NoPuffsNoPlumes
9H765CA0K823Q7
15JUL19:12:03:22
HDGB
?
2016 FORD TransitT350 5900
0
NoPuffsNoPlumes
8L6990A1K625H7
21JUL19:13:40:49
Table 3-18. Refueling Plume Results for HDGV3s
Vehicle
Class
ORVR
Assignment
Vehicle Description
Phase 1
Emissions
Summary
Phase 2
Emissions
Summary
VehID
i_Depart_MTN
(LinkTime)
HDGV3
no ORVR
1990 JAMB Fordlncomplete 9000
H
4B537SB1H808O2
23JUL19:14:21:49
HDGV3
no ORVR
1993 CON Fordlncomplete 10300
H
F02C62JFCF02C6
23JUL19:13:14:11
HDGV3
pre-ORVR
2002 FORD E3 5 0SuperDuty2WD 7100
0
NoPuffsNoPlumes
8L765CB7H263P4
08JUL19:08:43:56
HDGV3
?
2006 CHEVROLET 35002WD 6700
H
ContinuousPlume
17Z1582QE5W244
23JUL19:13:20:30
HDGV3
?
2015 GMC Savana3500 7660
H
ContinuousPlume
9D005C201060S4
15JUL19:08:34:25
3-70
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 3-19. Refueling Plume Results for HDGV4s
Vehicle
Class
ORVR
Assignment
Vehicle Description
Phase 1
Emissions
Summary
Phase 2
Emissions
Summary
VehID
i_Depart_MTN
(LinkTime)
HDGV4
no ORVR
1997 FLE Fordlncomplete 11000
H
W17S37VWSW17S3
18JUL19:14:25:47
HDGV4
no ORVR
2007 WINN Fordlncomplete 16300
H
ContinuousPlume
18D610AZD76201
12JUL19:09:13:46
HDGV4
no ORVR
2011 FORD Fordlncomplete missing
H
ContinuousPlume
T96L5 6QTLT96L5
18JUL19:10:20:45
HDGV4
no ORVR
2016 THOR Fordlncomplete 9700
H
ContinuousPlume
7QK6FK8767QK6F
22JUL19:11:41:48
3-71
-------
High Evaporative Emissions Investigation Field Study
Final Report
4.0 Thornton: Introduction to Liquid Refueling Emissions Study
ERG collected field data at the Thornton, Colorado, Costco Wholesale gas station with
the purpose of characterizing liquid gasoline spills made by customers as they refueled their
personal vehicles. This activity was performed in July 2019 as part of EPA Work Assignment 2-
23 under Contract EP-C-17-011 with the permission and assistance of Costco local, regional, and
national management. The goal of the study was to quantify the prevalence and magnitude of
private vehicle refueling liquid spills and their association with customer behavior and other
potential factors.
With one technician at the station at a given time, four technicians investigated the
occurrence of gasoline spills between 8 a.m. and 4 p.m. for fourteen days between July 7 and 23.
By the end of the study period, the technicians had observed 1,227 refueling events and had
identified 153 spills of various sizes. In addition to spill characteristics, the technicians recorded
factors and behaviors that could possibly affect the likelihood and severity of spills. These
variables include the pump number, location of fuel fill door with respect to fuel pump, nozzle
orientation, number of extra clicks (attempts to top off the vehicle tank), and idling state.
Monitoring for gasoline station refueling behavior and clicks, spills, and spitbacks was
conducted at the Costco Wholesale Thornton Colorado gas station, 16375 Washington Street,
Thornton, Colorado 80023. This gas station was open for business Monday through Friday from
6 a.m. to 9 p.m. and on Saturday and Sunday from 7 a.m. to 7 p.m.
A Google Maps photograph of the Costco Thornton site is shown in Figure 4-1. The
approximate dimension of the gas station canopy is 32 by 86 ft. Traffic is allowed to flow into
the gas station one way from the south as shown by the arrows on the pavement in the
photograph. The Thornton gas station is made up of 12 gas pumps on three islands. The layout of
the islands is shown in Figure 4-2. The direction of traffic, which flows north, is indicated by the
arrows in Figure 4-2. Customers are not permitted to enter the station from the exit driveway.
The arrangement of the islands and pumps in the figure corresponds to the same arrangement in
the Google Maps photograph in Figure 4-1.
For the data collected at the Costco Thornton station, the PG environmental technician
was not located at a single pump or island for testing but instead moved around the site to
monitor refueling behavior and liquid gasoline emissions.
4-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 4-1. Costco Thornton Site Used for Spills Evaluations
Figure 4-2. Diagram of Costco Thornton Gas Station Islands and Pumps
T
2 1
4 3
t T
6 5
8 7
t T
10
9
12
11
T
4-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
5.0 Thornton: Data Collection Methods
Each technician collected on-site data using two strategies: 1) sequential and 2) station-
wide. To do so, they recorded their observations in data packets containing: 1) white-page
datasheets and 2) pink-page datasheets. The technician instructions are given in Appendix D.
The technicians used the white-page datasheets to follow the vehicles sequentially for each
vehicle's entire refueling activity. After the conclusion of one vehicle's refueling activity, the
technician was instructed to select the next vehicle to be monitored by looking for the next
customer that exited their vehicle to refuel, regardless of whether there were other customers
already refueling. The white-page procedure was used to routinely monitor randomly selected
vehicle refueling; however, if a liquid spill anywhere in the station caused a station-wide
commotion, the technician was instructed to abort the current white-page data entry so that they
could investigate the source of the incident using the pink-page datasheets. This was an effort to
gather station-wide data on major spills and spitbacks.
After the completion of the onsite data collection, the technicians transcribed their own
written records into Excel spreadsheets. The spreadsheets were processed before analysis began,
while the paper datasheet packets were kept for back-up. The four technicians made a total of
1,193 sequential (white-page) observations and 34 station-wide (pink-page) observations. The
analysis described below covers both the 1,193 sequential observations and the 34 station-wide
observations.
5-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
6.0 Thornton: Data Processing and Database Assembly
Table 6-1 shows the datasheet format and two transcribed entry samples. The first entry
is incomplete. In the Text data field, the technician made a note that they moved to observe a
station-wide event. For interrupted entries, or where customers did not refuel their vehicle, the
entire observation was removed from further analysis. Of the 1,193 sequential observations, 22
were omitted because of station-wide event interruptions or non-refueling vehicles. This results
in 1,171 complete sequential observations over the fourteen days of data collection.
The second entry is complete except for the Spill Source, Spill Size, and Attendant
Action data fields. In instances where data fields were missing, the paper datasheet entries were
checked for transcription omissions. If the transcription was the source of the error, then the
Excel spreadsheet was corrected. If both the paper datasheet and the Excel spreadsheet had the
same fields missing, then the entry was used for further analysis except where blank. These
blank data fields were noted as missing for subsequent count and percentage calculations. Other
minor operations were coded within a SAS environment30 to account for typos, inconsistent
labeling conventions between technicians, and text comments.
30 P:\EPA_RefuelingEmissions_WA2-23\Summer2019\Analysis\CSS\Analysis\Analysis_CSS.sas
6-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 6-1. Two Transcribed Datasheet Entries.
Date
Day of
Week
Pump
Number
License
Make
Model
Fueling
Side
(Near, Far)
Nozzle
Orientation:
RightSideUp
UpSideDown
SideWays
Number of Extra Clicks
(0=auto shut-off only)
Nozzle
Hang-Up
Time
(hh:mm:ss)
Spitback?
(Yes, No)
Spills
Attendant Action
(None, Kitty, Spray, Cone)
Idling
(Yes, No)
Text:
If there was an
event, tell about it in
the cells below this
heading, i.e., Use
the full width of the
page to enter your
text.
State
Plate
Source:
Fill neck
Under car
Size:
Nickel
Tennis
Grapefruit
Bucket
07/15/2019
Mon
9
CO
Ford
Fusion
N
RSU
went to pink pages
8-Jul
Mon
5
CO
Subaru
Outback
N
RSU
1
3:21:22
N
N
6-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
7.0 Thornton: Results
For the sequential observations, as shown in Table 7-1, 10.3% of vehicles spilled
gasoline, though most spills were small. The technicians categorized spills using representative
diameter sizes: none, nickel-sized, tennis-ball-sized, grapefruit-sized, but if a spill was
substantially larger than the diameter of a grapefruit, it was called bucket-sized to reflect the fuel
volume of a small bucket (diameter of bucket and larger). As shown in Table 7-2, only 27 of the
120 gasoline spills were larger than nickel-sized, and a total of 6 bucket-sized spills occurred. In
addition, three of the spills were also spitbacks as shown in Table 7-3. Spitbacks are violent
events in which a vehicle's fuel tank expels gasoline during an otherwise routine refueling.
Spitbacks are sometimes attributed (by consumers) to faulty fueling nozzles, malfunctioning
emission control systems, or user error, but the root cause of such events is not fully understood.
Spill occurrence, spill size, and spitback frequencies are shown in Table 7-1, Table 7-2, and
Table 7-3, respectively.
Table 7-1. Spill Occurrence Frequency for Sequential Observations
Spill Occurred31
Frequency
Percent
No
11 )4o
v/ "
Yes
120
10.3
Total
1166
100
Table 7-2. Spill Size Frequency for Sequential Observations
Spill Size32
Frequency
Percent
None
1046
89.8
Nickel
92
7.9
Tennis Ball
16
1.4
Grapefruit
5
0.4
Bucket
6
0.5
Total
1165
100
Table 7-3. Spitback Frequency for Sequential Observations
Spitback33
Frequency
Percent
No
1 107
~
Yes
3
0.3
Total
1170
100
31 5 observations were missing an indication of Spill Occurrence.
32 6 observations were missing an indication of Spill Size.
33 1 observation was missing an indication of Spitback.
7-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
The station-wide observations include information on 33 additional spills, including 22
bucket-sized spills and 12 spitbacks. There was one station-wide observation in which the
customer did not spill gasoline but idled while refueling.
The gas station had three islands with twelve different pumps, labeled by different pump
numbers, as shown in Figure 4-2. Table 7-4 shows the frequency in which each pump was
visited. For the white sheet data, the technicians observed Pump 7 over nine times more
frequently than Pump 4. The customer's fuel fill door location, where the customer entered the
gas station, and the ease of access of specific pumps may have affected the frequency in which
each pump was visited.
Table 7-4. Pump Number Frequency for Sequential Observations
Pump Number34
Frequency
Percent
1
150
12.9
2
26
2.2
3
75
6.4
4
19
1.6
5
128
11.0
6
116
9.9
7
174
14.9
8
36
3.1
9
147
12.6
10
134
11.5
11
89
7.6
12
73
6.3
Total
1167
100
The overall rates of the various fueling sides, nozzle orientations, and extra clicks are
detailed in Table 7-5 to Table 7-7. Each of these variables was thought to possibly have some
effect on the likelihood of spills. Typically, customers fueled from the near side, i.e., most
customers parked their vehicle so that their fuel fill door faced the pump (89.2%), oriented their
nozzles right-side up (96.2%), added no extra clicks (67.5%), and did not idle (99.9%). Notably,
almost one-third of customers added at least one extra click after the automatic shut-off of their
nozzle. The relatively large proportion of customers attempting to add more gasoline by using
34 4 observations were missing an indication of Pump Number.
7-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
extra clicks suggests that customers do not believe that the nozzle's automatic shut-off produces
full tanks.
Table 7-5. Fueling Side Frequency for Sequential Observations
Fueling Side
Frequency
Percent
Near
1D44
v/ ;
Far
127
10.8
Total
1171
100
Table 7-6. Nozzle Orientation Frequency for Sequential Observations
Nozzle Orientation
Frequency
Percent
Right-side Up
1127
96.2
Sideways
40
3.4
Upside Down
4
0.3
Total
1171
100
Table 7-7. Extra Clicks Frequency for Sequential Observations
Extra Clicks35
Frequency
Percent
0
787
67.5
1
148
12.7
2
63
5.4
3
53
4.6
4-9
84
7.2
10-18
31
2.7
Total
1166
100
Table 7-8 shows the idling characteristics of the Costco Thornton vehicles at the gas
station in the dataset. The technicians were not instructed to note the idling state until the second
day of the study, leading to the 95 missing values. The National Fire Prevention Association
code 30A does not allow the idling of vehicles and equipment during fueling. The fraction of
vehicles idling at Costco gas stations may be lower than at stations of other companies because
Costco employs attendants who circulate among the customers. If needed, the gas station
attendants remind customers to turn off their engines while refueling.
35 5 observations has a missing value for number of Extra Clicks.
7-3
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 7-8. Idling Frequency for Sequential Observations
Idling36
Frequency
Percent
No
1075
99.9
Yes
1
0.1
Total
1076
100
36 95 observations were missing an indication of Idling.
7-4
-------
High Evaporative Emissions Investigation Field Study
Final Report
8.0 Thornton: Analysis
The factors and behaviors that were associated with the likelihood and severity of spills
were reviewed using logistic regression procedures in SAS.
Fueling from the far side, where the customer drags the fueling hose over or around their
vehicle to refuel, was examined for a correlation with spill occurrence. The Thornton gas station
had 14-foot fuel pump hoses so that customers could easily refuel from the far side. Being able to
refuel from either side increases the efficiency of refueling traffic, but it was not known whether
fueling from the far side increases spill occurrence. Table 8-1 shows the percentage of total spills
versus the percentage of all refuelings that occurred for the near and far fueling sides. 11.7% of
all spills occurred when customers refueled from the far side, while 10.8% of all refuelings
occurred from the far side. The difference between each side's spill and refuel frequency is
within 1%. With the small variation and a P-value of 0.75, we cannot claim that the likelihood of
spills is affected by the fueling side (we fail to reject the null hypothesis). The fueling side has
minimal correlation with spill occurrence.
Table 8-1. Fueling Side Frequency v. Total Spills and Refuelings
for Sequential Observations
Fueling Side
Spills
Refuelings
Number
Percent
Number
Percent
Near
106
88.3
1044
89.2
Far
14
11. 7
127
10.8
Total
120
100
1171
100
The nozzle orientation at insertion was another candidate for influencing the spill
occurrence. A total of 44 customers refueled with sideways nozzles or upside-down nozzles. The
percentage of total spills are minimally different from the percentage of all refuelings that
occurred for each nozzle orientation, as shown in Table 8-2. Like the results for fueling side,
each nozzle orientation's spill and refueling percentages were within 1% of each other. Upside-
down nozzles were associated with one spill (0.8% of total spills) out of four refuelings (0.3% of
total refuelings). Although the difference is less than 1%, the magnitude of the spill percentage is
more than double that of the refueling percentage. However, with the small number of upside-
down refuelings, a single decrease in its number of spills would have resulted in zero spills for
the orientation. With the small variation and a P-value of 0.55, we cannot claim that the
likelihood of spills is affected by the nozzle orientation (we fail to reject the null hypothesis).
The nozzle orientation has minimal correlation with spill occurrence.
8-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 8-2. Nozzle Orientation Frequency v. Total Spills and Refuelings
for Sequential Observations
Spills
Refuelings
Nozzle Orientation
Number
Percent
Number
Percent
Right Side Up
116
96. 7
1127
96.2
Side Wavs
3
2.5
40
3.4
Up Side Down
1
0.8
4
0.3
Total
120
100
1171
100
The number of extra clicks is the key factor that is positively correlated with both the
occurrence and size of spills. Extra clicks occur when customers attempt to add additional
gasoline to their tank after the automatic shut-off of their pump nozzle. An automatic shut-off
with no further attempt to add gasoline is considered as "no extra clicks," while each additional
attempt to add gasoline results in one extra click.
In this study, customers who added any number of extra clicks spilled more frequently
than those who accepted an automatic shut-off Table 8-4 show the number of spills and
refuelings as a function of number of extra clicks. The last column shows the spill rate, which is
the percent of refuelings that had a spill. The spill rate is lowest (8.4%) for zero extra clicks and
highest (20.6%) for two extra clicks. Table 8-4 shows the relative occurrences of spills and
refuelings. That table shows that when two extra clicks were added, 10.9% of all spills occurred,
but only 5.4% of all customers added two extra clicks.
Table 8-3. Spill Rate v. Extra Clicks
for Sequential Observations
Extra Clicks37
Number
of
Spills
Number
of
Refuelings
Spill Rate
(%)
0
66
787
8.4%
1
18
148
12.2%
2
13
63
20.6%
3
7
53
13.2%
4-9
11
84
13.1%
10-18
4
31
12.9%
Total
119
1166
10.2%
37 5 observations had a missing value for number of Extra Clicks.
8-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 8-4. Extra-Clicks Frequency v. Total Spills and Refuelings
for Sequential Observations
S
3ills
Refuelings
Extra Clicks38
Number
Percent
of Total
Number
Percent
of Total
0
66
55.5
787
67.5
1
18
15.1
148
12.7
2
13
10.9
63
5.4
3
7
5.9
53
4.6
4-9
11
9.2
84
7.2
10-18
4
3.4
31
2.7
Total
119
100
1166
100
The output of the logistic regression relating extra clicks to spill occurrence is shown in
Figure 8-1. A good fit is established when the spill occurrence is modeled as a function of the
natural logarithm of extra clicks, resulting in a P-value of 0.0089. The predicted spill likelihood
is 8.9% at zero extra clicks and rises to 20.2% at 18 extra clicks. Because the P-value < 0.05, we
conclude that the number of extra clicks has a significant effect on the probability of a spill
occurring (we reject the null hypothesis that extra clicks have no effect on spill occurrence).
Table 8-5 relates the number of extra clicks by the spill size. When no extra clicks were
added, any spills that occurred were most likely to be nickel-sized. Furthermore, the spills of
customers who added any number of extra clicks were on average larger than the spills of
customers who accepted an automatic shut-off The correlation is most apparent for the category
of bucket-sized spills. There was a total of 6 bucket-sized spills within the 1,171 sequential
observations. Of the 785 refuelings with no extra clicks, only one bucket-sized spill occurred.
However, 2 bucket-sized spills occurred for the 31 refuelings in the range of 10 to 18 extra
clicks. From the sequential observations, the likelihood of a bucket-sized spill is about 50 times
greater for customers adding 10 to 18 extra clicks when compared to customers accepting an
automatic nozzle shut-off
38 5 observations had a missing value for number of Extra Clicks.
8-3
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 8-1. Logistic Regression Total Spill Probability vs. Extra Clicks
Extra Clicks
Spill Likelihood Estimated Probability Lower 95% Confidence Limit
Upper 95% Confidence Limit
/EPA_RefuelingEmissions_WA2-23/5ummer2019/Analysis/C5S/AnalysisMnalysis_CSS.sas 12NOV19 17:34
Table 8-5, Extra Clicks Frequency by Spill Size
for Sequential Observations
Spill Size
Extra Clicks39
None
Nickel
Tennis Ball
Grapefruit
Bucket
Total
0
719
55
8
2
1
785
1
129
14
2
1
1
147
2
50
7
3
1
1
62
3
45
6
1
0
0
52
4-9
73
8
1
1
1
84
10-18
27
1
1
0
2
31
Total
1043
91
16
5
6
1161
39 10 observations were missing an indication of number of Extra Clicks, of Spill Size, or of both.
8-4
-------
High Evaporative Emissions Investigation Field Study
Final Report
The result of a logistic regression relating extra clicks to the size of spills confirms the
trends seen in Table 8-5. The effect of the natural logarithm of extra clicks and the interaction of
size and the natural logarithm of extra clicks on the cumulative size response produce P-values
of <0.0001 and 0.015, respectively. Thus, the logistic regression supports the notion that extra
clicks affect the spill size likelihood.
The output of the logistic regression relating extra clicks to the spill size is shown in
Figure 8-2. In the figure, the spill sizes are modeled cumulatively as indicated by the legend. The
increased spill likelihood at large numbers of extra clicks is primarily due to the emergence of
grapefruit and bucket-sized spills. The model reconfirms that bucket-sized spills are very
infrequent at no extra clicks (0.15%) but become far more likely by 18 extra clicks (8.0%).
Extra Clicks
Cumulative Size — all spills — bucket-sized
— grapefruit or larger "~ tennis ball or larger
VEPA_RefiielingEmissions_WA2-23/5ummer2019/Analysis/CS5/Analysis/Analysis_CSS.sas 12NOV19 16:38
Human error may be why extra clicks result in more spills. A typical nozzle will
automatically shut off when gasoline reaches its sensing port near the tip of its nozzle spout.
When an automatic shut-off occurs, disregarding false click-offs, the nozzle has functioned
properly. While hanging up the nozzle, a small spill may still occur due to gasoline coating the
8-5
-------
High Evaporative Emissions Investigation Field Study
Final Report
nozzle outlet pipe, but large spills should not arise unless the nozzle lever is accidentally pressed
by the customer. If a nozzle fails to automatically shut off, then a spitback may occur. However,
spitback incidents are rare and are not limited to 0 extra clicks. Within the sequential
observations, one of the 3 spitbacks occurred after 2 extra clicks, and only one bucket-sized
spitback was recorded.
When customers pay using a credit or debit card, the automatic shut-off does not lock out
additional clicks. The gas pumps at the Costco Thornton station accept only cards for gasoline
purchases. Some customers who disregard the automatic shut-off partially raise their nozzle
spout to continue to refuel. Gasoline must reach the new level of the sensing port before another
shut-off click can be triggered. Modern-day nozzles can have flow rates of up to 10
gallons/minute and depending on the height in which the nozzle is raised, different types of spills
may occur. If the subsequent automatic shut-off does not trigger in time, a spitback may take
place. If the nozzle spout is too high, then the customer's stream may simply miss the fill pipe.
With additional sources of human error, larger spills become more common.
Station-wide (pink-page) observations were made to help confirm sequential results for
bucket-sized spills and spitbacks, which might not have been sampled well enough by the
sequential method. The station-wide observations contained an additional 22 bucket-sized spills
and 12 spitbacks. Station-wide observations were made only when significant incidents were
noticed, so the spills and extra clicks typically occurred prior to the technician's presence at the
pump having the spill or spitback. Consequently, only six of the bucket-sized spill observations
contained data on the number of extra clicks. A larger study would enable a more thorough
assessment of the impact of extra clicks on large gasoline spills.
An aggregate count of refuelings only when the technician was on duty is needed to
assess whether the sequential refueling observations adequately quantified the fraction of
refuelings that produced bucket-sized spills and spitbacks. Only refuelings that occurred while
the technician was on duty could result in a station-wide observation. The refuelings were
determined from the Costco transaction data. Any transactions that occurred prior to the first
observation, during breaks, or after the last observation of each day are not included in the
aggregate refuelings. The frequency of elapsed times between adjacent refueling observations is
given in Figure 8-3. The distribution in the figure shows a clear change in characteristics at an
elapsed time of 15 minutes. Times shorter than 15 minutes have a smooth distribution; times
longer than 15 minutes are rare and widely scattered. Thus, we designated elapsed times of
fifteen or more minutes as technician breaks. Over the fourteen-day study, 9,038 refuelings
occurred while a technician was on duty.
8-6
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure 8-3, Elapsed Times between Adjacent Refueling Observations
CUM.
CUM.
observation gap MIDPOINT
FREQ.
FREQ.
PCT.
PCT.
0:00:30
15
15
1.28
1.28
0:01:30
25
40
2.13
3.41
0:02:30
120
160
10.22
13.63
0:03:30
350
510
29.81
43.44
0:04:30
300
810
25.55
68.99
0:05:30
135
945
11.50
80.49
0:06:30
81
1026
6.90
87.39
0:0730
48
1074
4.09
91.48
0:08:30
26
1100
2.21
93.70
0:0930
18
1118
1.53
95.23
0:10:30
hi
14
1132
1.19
96.42
0:11:30
10
1142
0.85
9727
0:12:30
i
4
1146
0.34
97.61
0:13:30
3
1149
0.26
97.87
0:14:30
1
1150
0.09
97.96
0:15:30
0
1150
0.00
97.96
0:16:30
1
1151
0.09
98.04
0:17:30
1
1152
0.09
98.13
0:18:30
1
1153
0.09
98.21
0:19:30
2
1155
0.17
98.38
0:20:30
2
1157
0.17
98.55
0:21:30
1
1158
0.09
98.64
0:22:30
1
1159
0.09
98.72
0:23:30
2
1161
0.17
98.89
0:24:30
2
1163
0.17
99.06
0:25:30
0
1163
0.00
99.06
026:30
0
1163
0.00
99.06
0:27:30
0
1163
0.00
99.06
0:28:30
1
1164
0.09
99.15
0:29:30
0
1164
0.00
99.15
0:30:30
i
3
1167
0.26
99.40
0:31:30
0
1167
0.00
99.40
032:30
1
1168
0.09
99.49
0:33:30
0
1168
0.00
99.49
0:34:30
1
1169
0.09
99.57
035:30
0
1169
0.00
99.57
0:36:30
0
1169
0.00
99.57
0:37:30
0
1169
0.00
99.57
0:38:30
0
1169
0.00
99.57
0:39:30
1
1170
0.09
99.66
0:40:30
0
1170
0.00
99.66
0:41:30
0
1170
0.00
99.66
0:42:30
1
1171
0.09
99.74
0:43:30
0
1171
0.00
99.74
0:44:30
0
1171
0.00
99.74
0:45:30
2
1173
0.17
99.91
0:46:30
1
1174
0.09
100.00
0 100 200 300 400
FREQUENCY
/EPA_RefuelingEmissions_WA2-23/Summer2019/Ana|ysis/CSS/Anaysis/Pink Analysis Transaction Data.sas 07OCT19 12:48
During the sequential observations, bucket-sized spills occurred at a rate of 0.5% (see
Table 7-2) and spitbacks occurred at a rate of 0.3% (see Table 7-3). The aggregate frequencies of
bucket-sized spills and spitbacks, based on white and pink-page data, are given in Table 8-6 and
Table 8-7, respectively. The frequencies are composed of both sequential (white-page) and
station-wide (pink-page) observations. The observed resultant rates of 0.31% for bucket-sized
spills and 0.17% for spitbacks are lower than their true probabilities because there could have
been cases where large spills were not noticed and therefore were not recorded. A portion of the
station-wide observations were recorded after the attendant notified the on-duty technician of a
spill. The gas station attendants are more likely to encounter station-wide events than the
technicians. The attendants do not need to record refueling observations, and some customers
who experience large spills or spitbacks will ask an attendant for help. Given that some station-
wide events go unnoticed, the real probability of bucket-sized spills and spitbacks at the Costco
gas station in Thornton, Colorado should be considered at least 0.31% and 0.17%, respectively.
Therefore, we regard these station-wide rates for bucket-sized spills (0.31%) and spitbacks
(0.17%) to be in substantial agreement with the sequential rates of 0.5% and 0.3%, respectively.
8-7
-------
High Evaporative Emissions Investigation Field Study
Final Report
Table 8-6. Bucket-Sized Spill Frequency for Station-Wide Refuelings
Bucket-sized Spill
Frequency
Percent
Confirmed
28
0.31
Other Refuelings
9010
99.69
9038
100
Table 8-7. Spitback Frequency for Station-Wide Refuelings
Spitback
Frequency
Percent
Confirmed
15
0.17
Other Refuelings
9023
99.83
9038
100
-------
High Evaporative Emissions Investigation Field Study
Final Report
9.0 Concluding Thoughts for Future Consideration
During collection and analysis of the Arvada and Thornton refueling data, we had ideas
for improvements that can be made on the data collection methods and possible follow-on
studies for characterizing evaporative emissions control systems in the field:
1) The Rebellion Photonics Gas Cloud Imaging camera used at Arvada was effective at
detecting refueling emissions (puffs and plumes) in the Enhanced MidWave videos, which are
created from infrared absorption in the 3.2 to 3.5 |im region. Rebellion uses special processing of
7.5 to 14 |im infrared absorption data, in addition to the 3.2 to 3.5 |im data, to quantify emissions
for its clients. However, after examining trial processing of selected reference vehicle videos, we
concluded that the infrared quantitative processing that Rebellion offered would not be
successful in quantifying the emissions for each refueling event with the accuracy desired. This
was a consequence of vehicle and people movement in the background, inconsistent background
illumination, and low emissions concentrations for properly functioning control systems.
Since the Rebellion approach would not be likely to quantify refueling emissions, a less
expensive approach might be possible. A non-quantification study using infrared video to image
refueling vapor emissions using a forward-looking infrared (FLIR) video camera might be
considered. This type of camera is more common than the GCI camera and operates in the same
3.2 to 3.5 |im infrared region. We have not investigated the sensitivity of FLIR cameras for this
study to determine if the technique would be capable of imaging refueling emissions. However,
ERG has experience with FLIR cameras in other areas of emissions study.
2) If quantification of refueling emissions vs. time is desired, a brute-force, lower-tech
method might be used. For example, with a cooperative gas station owner, a portable SHED
(PSHED) with sealable fabric entry and exit doors might be set up at one fuel pump. A
hydrocarbon analyzer could measure HC concentration as the owner refueled his vehicle through
an opening in the side of the PSHED. The concentration vs. time profile of each refueling event
could be used to calculate the emission mass. A stratified, random plan based on model year
might be used to sample vehicle candidates. Drivers might be enticed to participate by offering
free gasoline. With efficient logistics, the refueling emissions, including puffs and plumes, could
be measured rapidly.
3) One unanswered question in this study was the determination of the causes of
continuous plumes from ORVR vehicles. A study such as this one can effectively identify those
vehicles by using the infrared camera to screen many vehicles as they refuel. With a gas station
9-1
-------
High Evaporative Emissions Investigation Field Study
Final Report
owner's assistance, the owners of identified vehicles might be approached to participate in a
follow-on study of laboratory measurement of evaporative emissions and vehicle examination.
Vehicle owners might be offered a substantial incentive and use of a rental car for the few days
while their vehicle is being tested.
4) In a liquid-spill study like at Thornton, the size of spills should be more accurately
measured. This could be simply done by estimating the diameter or area of the spilled gasoline
on the pavement. Also, in this study, gasoline leaking from vehicles was determined by
examining the pavement where vehicles were sitting after they drove away. Since many vehicles
were dripping water from a nearby car wash, technicians had a difficult time determining
whether drips were gasoline or water without touching or smelling he drips. A handheld
electronic HC detector could easily solve that problem.
5) Fuel pump nozzle manufacturers have developed new "dripless" nozzles to help
reduce liquid gasoline spills. Since a single technician with a clipboard or iPad is all that is
needed to collect gas station behavior data, a comparative study to quantify the benefits of such
nozzles vs. conventional nozzles could be inexpensively done at a participating station if
different pump nozzles types were installed on different fuel pumps.
9-2
-------
High Evaporative Emissions Investigation Field Study Final Report
Appendix A
Arvada: Reference Vehicle Test Conditions and Results40
40 C:\Documents\EPA CanisterDegradation\WA2-23 (GasStnRebellion_MAR2019)\Data
QC\RefVehicleTests/RefVehLogsheets-190802.xlsx
-------
High Evaporative Emissions Investigation Field Study
Final Report
Date
Release
Start
Time
Video
Start
Time
Video Filename
(*.mp4)
Pump
Test
Condition
Number
Test
Condition
Name
Plume
Seen?
07/10/2019
14:19
14:19:14
viewer 1562789954316
9
1
GAS 100 L DOOR
1
07/08/2019
17:48
17:48:24
viewer 1562629703976
7
1
GAS 100 L DOOR
1
07/12/2019
08:06
08:06:32
viewer 1562940392482
11
1
GAS 100 L DOOR
1
07/16/2019
15:17
15:17:15
viewer 1563311834967
11
1
GAS 100 L DOOR
1
07/17/2019
08:06
08:05:56
viewer 1563372356322
6
1
GAS 100 L DOOR
1
07/10/2019
14:26
14:26:38
viewer 1562790397771
9
4
GAS 030 L DOOR
1
07/08/2019
17:54
17:54:30
viewer 1562630069759
7
4
GAS 030 L DOOR
1
07/12/2019
08:11
08:11:46
viewer 1562940706278
11
4
GAS 030 L DOOR
1
07/15/2019
19:45
19:45:40
viewer 1563241540391
9
4
GAS 030 L DOOR
1
07/16/2019
15:25
15:25:35
viewer 1563312335081
11
4
GAS 030 L DOOR
1
07/17/2019
08:10
08:10:43
viewer 1563372642877
6
4
GAS 030 L DOOR
1
07/10/2019
14:31
14:30:58
viewer 1562790658301
9
7
GAS 010 L DOOR
1
07/08/2019
17:59
17:59:50
viewer 1562630390229
7
7
GAS 010 L DOOR
1
07/12/2019
08:16
08:16:31
viewer 1562940990695
11
7
GAS 010 L DOOR
0
07/15/2019
19:37
19:37:36
viewer 1563241056222
9
7
GAS 010 L DOOR
0
07/16/2019
15:33
15:33:43
viewer 1563312822739
11
7
GAS 010 L DOOR
1
07/17/2019
08:15
08:15:39
viewer 1563372938525
6
7
GAS 010 L DOOR
1
07/10/2019
13:47
13:47:57
viewer 1562788077434
9
10
BUT 100 L DOOR
1
07/08/2019
17:28
17:28:04
viewer 1562628483786
7
10
BUT 100 L DOOR
1
07/12/2019
07:44
07:44:33
viewer 1562939072840
11
10
BUT 100 L DOOR
1
07/15/2019
19:06
19:05:55
viewer 1563239155450
9
10
BUT 100 L DOOR
1
07/16/2019
15:01
15:01:17
viewer 1563310877302
11
10
BUT 100 L DOOR
1
07/17/2019
07:47
07:47:22
viewer 1563371241757
6
10
BUT 100 L DOOR
1
07/10/2019
14:01
14:00:57
viewer 1562788857493
9
13
BUT 030 L DOOR
1
07/12/2019
07:54
07:54:04
viewer 1562939644496
11
13
BUT 030 L DOOR
1
07/15/2019
19:21
19:21:13
viewer 1563240072669
9
13
BUT 030 L DOOR
1
07/16/2019
15:06
15:05:53
viewer 1563311152661
11
13
BUT 030 L DOOR
1
07/17/2019
07:36
07:36:44
viewer 1563370604099
8
13
BUT 030 L DOOR
1
07/17/2019
07:52
07:52:46
viewer 1563371566031
6
13
BUT 030 L DOOR
1
07/10/2019
14:06
14:06:19
viewer 1562789178766
9
16
BUT 010 L DOOR
07/12/2019
07:59
07:59:14
viewer 1562939954357
11
16
BUT 010 L DOOR
1
07/15/2019
19:27
19:27:04
viewer 1563240423955
9
16
BUT 010 L DOOR
1
07/16/2019
15:10
15:10:36
viewer 1563311435617
11
16
BUT 010 L DOOR
1
07/17/2019
07:42
07:41:54
viewer 1563370913961
8
16
BUT 010 L DOOR
1
07/17/2019
07:57
07:57:48
viewer 1563371867750
6
16
BUT 010 L DOOR
1
A-l
-------
High Evaporative Emissions Investigation Field Study
Final Report
Date
Release
Start
Time
Video
Start
Time
Video Filename
(*.mp4)
Pump
Test
Condition
Number
Test
Condition
Name
Plume
Seen?
07/10/2019
14:21
14:21:23
viewer 1562790083372
9
2
GAS 100 TANK
1
07/08/2019
17:49
17:49:40
viewer 1562629780385
7
2
GAS 100 TANK
1
07/12/2019
08:08
08:07:58
viewer 1562940477564
11
2
GAS 100 TANK
1
07/16/2019
15:19
15:18:56
viewer 1563311936402
11
2
GAS 100 TANK
1
07/17/2019
08:07
08:07:17
viewer 1563372436538
6
2
GAS 100 TANK
1
07/08/2019
17:55
17:55:50
viewer 1562630150107
7
5
GAS 030 TANK
1
07/12/2019
08:13
08:13:19
viewer 1562940799371
11
5
GAS 030 TANK
1
07/15/2019
19:50
19:50:33
viewer 1563241833284
9
5
GAS 030 TANK
1
07/10/2019
14:28
14:28:11
viewer 1562790491332
9
5
GAS 030 TANK
07/16/2019
15:29
15:29:13
viewer 1563312553068
11
5
GAS 030 TANK
1
07/17/2019
08:12
08:12:13
viewer 1563372732501
6
5
GAS 030 TANK
1
07/08/2019
18:01
18:01:10
viewer 1562630469908
7
8
GAS 010 TANK
0
07/10/2019
14:32
14:32:16
viewer 1562790736246
9
8
GAS 010 TANK
0
07/12/2019
08:19
08:19:49
viewer 1562941189372
11
8
GAS 010 TANK
0
07/15/2019
19:41
19:41:40
viewer 1563241299668
9
8
GAS 010 TANK
0
07/16/2019
15:37
15:37:03
viewer 1563313022556
11
8
GAS 010 TANK
0
07/17/2019
08:17
08:17:02
viewer 1563373022474
6
8
GAS 010 TANK
1
07/10/2019
13:52
13:52:39
viewer 1562788358582
9
10
BUT 100 TANK
1
07/10/2019
14:14
14:13:58
viewer 1562789638113
9
11
BUT 100 TANK
1
07/12/2019
07:49
07:49:05
viewer 1562939344997
11
11
BUT 100 TANK
1
07/15/2019
19:08
19:08:00
viewer 1563239279843
9
11
BUT 100 TANK
1
07/16/2019
15:02
15:02:45
viewer 1563310965389
11
11
BUT 100 TANK
1
07/17/2019
07:49
07:49:47
viewer 1563371387319
6
11
BUT 100 TANK
1
07/10/2019
14:02
14:02:51
viewer 1562788970757
9
14
BUT 030 TANK
1
07/08/2019
17:38
17:38:36
viewer 1562629116104
7
14
BUT 030 TANK
1
07/12/2019
07:56
07:55:59
viewer 1562939758962
11
14
BUT 030 TANK
1
07/15/2019
19:23
19:23:19
viewer 1563240198529
9
14
BUT 030 TANK
1
07/16/2019
15:07
15:07:18
viewer 1563311238282
11
14
BUT 030 TANK
1
07/17/2019
07:38
07:38:07
viewer 1563370686983
8
14
BUT 030 TANK
1
07/17/2019
07:54
07:54:28
viewer 1563371667869
6
14
BUT 030 TANK
1
07/10/2019
14:07
14:07:48
viewer 1562789267991
9
16
BUT 010 TANK
0
07/12/2019
08:00
08:00:38
viewer 1562940038239
11
17
BUT 010 TANK
0
07/17/2019
07:43
07:43:02
viewer 1563370982363
8
17
BUT 010 TANK
0
07/17/2019
07:59
07:59:19
viewer 1563371958843
6
17
BUT 010 TANK
0
07/15/2019
19:29
19:29:26
viewer 1563240566114
9
17
BUT 010 TANK
1
07/16/2019
15:12
15:12:43
viewer 1563311563361
11
17
BUT 010 TANK
1
A-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
Date
Release
Start
Time
Video
Start
Time
Video Filename
(*.mp4)
Pump
Test
Condition
Number
Test
Condition
Name
Plume
Seen?
07/10/2019
14:23
14:22:58
viewer 1562790177999
9
3
GAS 100 R DOOR
1
07/08/2019
17:51
17:51:47
viewer 1562629906863
7
3
GAS 100 R DOOR
1
07/12/2019
08:09
08:09:17
viewer 1562940557110
11
3
GAS 100 R DOOR
1
07/16/2019
15:20
15:20:49
viewer 1563312048931
11
3
GAS 100 R DOOR
1
07/17/2019
08:08
08:08:43
viewer 1563372523491
6
3
GAS 100 R DOOR
1
07/10/2019
14:29
14:29:28
viewer 1562790568210
9
6
GAS 030 R DOOR
1
07/08/2019
17:57
17:57:07
viewer 1562630226650
7
6
GAS 030 R DOOR
1
07/12/2019
08:14
08:14:48
viewer 1562940888125
11
6
GAS 030 R DOOR
1
07/15/2019
20:03
20:02:30
viewer 1563242549876
9
6
GAS 030 R DOOR
1
07/16/2019
15:31
15:31:14
viewer 1563312673922
11
6
GAS 030 R DOOR
1
07/17/2019
08:14
08:13:56
viewer 1563372835553
6
6
GAS 030 R DOOR
1
07/10/2019
14:35
14:35:35
viewer 1562790934930
9
9
GAS 010 R DOOR
1
07/08/2019
18:02
18:02:30
viewer 1562630550322
7
9
GAS 010 R DOOR
1
07/12/2019
08:21
08:21:12
viewer 1562941272057
11
9
GAS 010 R DOOR
1
07/15/2019
19:43
19:43:36
viewer 1563241415717
9
9
GAS 010 R DOOR
1
07/16/2019
15:40
15:40:49
viewer 1563313248914
11
9
GAS 010 R DOOR
1
07/17/2019
08:18
08:18:28
viewer 1563373108493
6
9
GAS 010 R DOOR
1
07/10/2019
13:56
13:56:35
viewer 1562788595369
9
12
BUT 100 R DOOR
1
07/08/2019
17:31
17:31:46
viewer 1562628705810
7
12
BUT 100 R DOOR
1
07/12/2019
07:52
07:51:57
viewer 1562939516569
11
12
BUT 100 R DOOR
1
07/15/2019
19:19
19:18:43
viewer 1563239923303
9
12
BUT 100 R DOOR
1
07/16/2019
15:04
15:04:30
viewer 1563311070179
11
12
BUT 100 R DOOR
1
07/17/2019
07:34
07:34:37
viewer 1563370476637
8
12
BUT 100 R DOOR
1
07/17/2019
07:51
07:51:23
viewer 1563371482749
6
12
BUT 100 R DOOR
1
07/10/2019
14:04
14:04:38
viewer 1562789078266
9
15
BUT 030 R DOOR
1
07/08/2019
17:39
17:39:47
viewer 1562629186708
7
15
BUT 030 R DOOR
1
07/12/2019
07:57
07:57:37
viewer 1562939856526
11
15
BUT 030 R DOOR
1
07/15/2019
19:25
19:25:19
viewer 1563240319384
9
15
BUT 030 R DOOR
1
07/16/2019
15:09
15:08:57
viewer 1563311336784
11
15
BUT 030 R DOOR
1
07/17/2019
07:39
07:39:31
viewer 1563370771134
8
15
BUT 030 R DOOR
1
07/17/2019
07:56
07:55:46
viewer 1563371746211
6
15
BUT 030 R DOOR
1
07/17/2019
08:00
08:00:45
viewer 1563372045396
6
18
BUT 010 R DOOR
07/10/2019
14:10
14:10:28
viewer 1562789427901
9
18
BUT 010 R DOOR
1
07/12/2019
08:02
08:02:07
viewer 1562940127463
11
18
BUT 010 R DOOR
1
07/15/2019
19:34
19:33:46
viewer 1563240825974
9
18
BUT 010 R DOOR
1
07/16/2019
15:14
15:14:27
viewer 1563311667002
11
18
BUT 010 R DOOR
1
07/17/2019
07:44
07:44:24
viewer 1563371063711
8
18
BUT 010 R DOOR
1
A-3
-------
High Evaporative Emissions Investigation Field Study Final Report
Appendix B
Arvada: Phase 1 Enhanced MidWave Video Viewing instructions41
41 C:\Documents\EPA CanisterDegradation\WA2-23 (GasStnRebellion_MAR2019)\Data
QC/VideoQCinglnstr-190918 .docx
-------
High Evaporative Emissions Investigation Field Study
Final Report
Overview of the SharePoint QC Database
Each row in the SharePoint database contains information for one refueling event for one vehicle
including:
Variables (sortable) on the main screen:
Title: A code that is unique to each vehicle.
Year: Vehicle model year according to Colorado registration database via license plate.
Make: Vehicle make according to Colorado registration database via license plate.
Model: Vehicle model according to Colorado registration database via license plate.
EmptyWt: Vehicle empty weight according to Colorado registration database via license plate.
PumpID: The identifier of the pump at the Costco Arvada gas station.
Gallons: The volume of gasoline dispensed.
Arrive_MTN: DateTime that the vehicle arrived at the pump.
Additional variables on the drop-down list (accessed by clicking on a Title on the main
screen):
VehicleShort: Concatenation of Year, Make, Model, EmptyWt.
Short Vehicle Match: An indicator of the match between appearance/description of the vehicle.
EnhMW_#_Video: A hyperlink to a 30-second video. Up to 6 videos per refueling event.
*_Vid Defect: Technician's evaluation of video quality.
*_PlumeChar: Technician's evaluation of plume presence and other qualities.
*_PlumeTime: Technician's evaluation of when plume is first seen in the video.
'•^Vehicle: Technician's evaluation of vehicle motion.
*_HoseDistance: Technician's evaluation of gas pump hose placement.
Abridged Phase 1 Instructions Used by Technicians to Evaluate Videos:
Each row in the SharePoint database contains information for one refueling event for one vehicle
including: a variable telling if the observation's videos should be watched (Target), a pump
identifier (iPumpID), a vehicle description (i VehicleShort), multiple EnhMW video hyperlinks
(r EnhMW #). The rows should be sorted by the datetime of the credit card authorization
timestamp (c_CCStart).
View only videos on rows that are marked as Targets. If Target is blank, do not watch any videos
on that row. Either the videos have already been watched or some portion of the vehicle
description is unknown.
To allow recording of EnhMW information, the SharePoint database has initially empty
variables for each of a row's EnhMW videos for the following five major categories with a drop-
down menu for each to allow selection of just one of the values within each category:
B-l
-------
High Evaporative Emissions Investigation Field Study
Final Report
Video Defects
Refueling plume characteristics
Video Second that the plume is first seen in the video
Vehicle Action
Hose Distance
Specific Phase 1 Instructions
1. Starting at the top of the SharePoint database, find the line with the first Target.
2. Note the iPumpID value:
If iPumpID = 5, 6, or 9, the target vehicle is the front pump.
If iPumpID = 7, 8, or 11, the target vehicle is the rear pump.
3. View r_EnhMW_l, which is the first video for the refueling event, by clicking on the
video filename.
4. As the video plays and looking only at the vehicle at the correct pump, select a single
response to the five categories using the drop-down menus corresponding to the video number (1
to 5). Watch the video multiple times if necessary.
5. Repeat the same procedure for all EnhMW videos on the line.
6. Go to the next line that is marked by Target and repeat the procedure.
Guidelines Used by Technicians to Evaluate Videos:
i_ShortVehicle_Match: Verify that the vehicle on the camera matches the short vehicle
description. Use google images to help identify cars. Use this primarily if you think that there is
no way that the vehicle in the video can match the registration database description. Usually
leave this field "Can't tell."
Match = Vehicle matches the description
Doesn't Match = Vehicle is very clearly different from the vehicle in description.
Can't Tell = If it is unclear (Default)
Video Defects:
+ = Good EnhMW video for entire 30s
• Video is black and white.
• Video is noisy. (Look for salt and pepper)
• Entire video is in motion, no still frames.
• No white screen present during video.
- Not an EnhMW video for some part of video
B-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
• Video is in color, not black and white.
• Video is not in enhanced infrared. The video is black and white but appears almost like
it's a video in negative light and isn't noisy.
F Frozen video
• Video freezes (still frame).
W White screen for some part of video.
• Video cuts to a white screen.
• This occurs when the Rebellion IR camera is calibrating, which occurred every 8
minutes.
X Video is completely useless since it has no plume info.
• If any of the video defects persist for more than 2/3 (20 seconds) of the video.
Refueling plume characteristics:
0 No plume visible
L Light plume
II Heavy plume
P Puddle on pavement after vehicle leaves with plume coming from it. (Takes priority.)
X Not possible to review for plume. (If video defect is X)
G Fueled something that isn't a car (Gas can. Boat, Jet ski, etc.)
S Plume is blown on-screen from an off-screen source.
Video Second that the plume is first seen in the video:
0 = no plume is ever seen in this video
# = (a number between 1 and 30) The second that a plume is first visible in the video.
Vehicle Action:
0 No vehicle is at this pump in this video.
A The targeted vehicle arrives at the pump in this video.
R The targeted vehicle is at the pump for the entire video.
D The targeted vehicle drives avvav from the pump in this video.
X The video has a vehicle different from the vehicle in the other videos assigned to this
refueling event.
Hose Distance:
N Near (the fuel till door is close to the pump)
F Far (the pump hose is stretched across the vehicle to get to the fuel till door)
B-3
-------
High Evaporative Emissions Investigation Field Study
Final Report
Appendix C
Arvada: Phase 2 Enhanced MidWave Video Viewing Instructions
-------
High Evaporative Emissions Investigation Field Study
Final Report
The Phase 1 video viewing found that some refueling events showed refueling plumes.
About 455 refueling events of confirmed ORVR vehicles had overall Phase 1 viewing results of
L, H, or P (low-density, high-density, or puddle). We need to re-view the Enhanced MidWave
videos of those events to determine the time trends of when plumes appeared. The reason is that
in many cases with plumes present, plumes appeared only briefly when the gas cap was removed
and/or when the fuel pump nozzle clicked off. We want to distinguish instances of that behavior
from the behavior when plumes are being produced during fuel-dispensing periods.
Each Enhanced MidWave video is 30s long. We divide each video into six 5s blocks and
evaluate each of the blocks in each video to produce a 6-digit string of characters. Assign a
single character to each 5s block.
During the Phase 1 video examination, we typically saw two types of refueling
emissions: 1) continuous plumes, and 2) puffs of emissions. Each type of emission can produce
codes, but use only one code for each 5s block, as described below.
Puff Codes - We define a "puff as a short-duration plume that seems to be associated
with a specific activity - especially gas cap removal, beginning of fuel flow, end of fuel flow at
gas pump nozzle click-off If you see a puff during these activities, use these codes for the 5 s
block that they occur in:
R = puff at gas cap Removal.
B = puff at Beginning of fuel flow, i.e., when the customer pulls the nozzle handle.
If you see a puff in a block when both gas cap removal and beginning of fuel
flow happen, and if you cannot tell whether the puff comes from R or B, then
record aB.
E = puff at End of fuel flow, i.e., when the nozzle clicks off.
For blocks other than those where a puff occurs, use the continuous plume codes
described below.
Continuous Plume Codes - As described above, during gas cap removal and nozzle
insertion and during and after nozzle click-off, if you see a puff of vapor, record the puff for the
5s block. But if a puff associated with those activities does not occur, use the codes for
continuous plumes - keeping in mind that continuous plumes can occur before and/or after those
three puff-associated activities. Here are the characters to be used to characterize a continuous
plume in each 5s block:
0 = no plume can be seen
1 = a low-contrast plume can be seen.
2 = a high-contrast plume can be seen.
X = the entire video screen is white, which indicates that the GCI camera is
calibrating.
C-l
-------
High Evaporative Emissions Investigation Field Study
Final Report
P = a plume is coming from a puddle of gasoline on the pavement. This might
occur at any time in the video. P takes priority over 1 and 2.
Record the continuous plume character that makes up the most time of the 5s block. For
example, if there are about 4 seconds of 0 and 1 second of 1, record 0 for the block.
A note about continuous plume evaluation. Contrast refers to the variation of light to dark
within the plume. Areas of very white and very black within the plume are high contrast and
would be assigned a 2. Areas with slight differences of gray indicate low contrast and would be
assigned a 2 - unless the plume is large - in which case it would be assigned a 2. In the Phase 1
viewing, we did not look so much at plume contrast but considered mainly plume size. The
problem is that the contrast is affected by the distance from the emission point, the speed and
direction of air movement, and the camera's view of the target vehicle and its surroundings.
Sometimes the view shows only a small portion of the surroundings, which means that large
plumes may not be in the video even if they exist off-screen. So, if a large, billowing plume is
seen over a large screen area, it can be a 2 even if the contrast within the plume is low. Also, if a
large, billowing plume cannot be seen because the vehicle is near the edge of the screen, but a
small, high-contrast stream of vapor is seen coming from the fuel fill door, this result can also be
a 2. If you are watching the video and you say to yourself: "Is that a plume? I think there
MIGHT be a plume there." That is a 1. A small wispy plume will also be a 1.
General Guidance - You can use the Phase 1 viewing results of a refueling event as
guidance, but do not think and do not try to make the Phase 2 time profile string to somehow
match the Phase 1 viewing results. The Phase 2 evaluation criteria are different from the Phase 1
viewing criteria.
Do not evaluate (just ignore) plumes from off-screen (formerly S), gas cans and off-road
equipment (formerly G). Evaluate only plumes that you judge are from the target vehicle.
Record a digit even though you can clearly see that the customer has not yet lifted the
nozzle from the pump or that he has already hung up the nozzle. We must have a 6-digit string
for every video.
You need to be reasonably certain that the plume you are evaluating is from the targeted
vehicle - not the other vehicle at the island. If you are not certain, use a ? as the character.
C-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
Appendix D
Thornton: Clicks, Spills, Spitback Data Collection Instructions
-------
High Evaporative Emissions Investigation Field Study Final Report
For the WHITE datasheets
Goal: Collect data to find out the fraction of refueling events that produce liquid gasoline spills.
Rule 1: Use your best printing when making these data entries. If you make a printing error or
you think that the entry may not be clear to someone else, correct it immediately. Make the
correction by carefully drawing a single horizontal line through the entry and re-writing the entry
above, below, or near the crossed-out entry. Never write over an erroneous character. Never
scribble over an erroneous entry. Never use an X to cross out an erroneous entry.
Step 1. Use the logsheet CSS_logsheet-190704.xlsx.
Step 2. Enter the Day of Week (Sun, Mon, Tue, ...) in the first cell of the data row.
Step 3. The next vehicle to be picked will be from any one of the 12 pumps at the station.
Pick the first vehicle where the driver gets out.
Step 4. Enter the Pump Number.
Step 5. Enter the license plate State using the rear license plate.
Step 6. Enter the rear Plate. Be very careful to clearly print the Plate. The license Plate is
a critical variable in this study. Clearly distinguish characters that are easily confused:
5 vs. S
2 vs. Z
D vs. Q vs. O vs. 0
I vs. 1
B vs. 8
U vs. V
G vs. C
7 vs. 9
If the plate is missing, enter the word "missing". If the plate is a paper temporary plate,
enter the word "temp".
Step 7. Enter the vehicle Make and Model. Sometimes it is not clear what the vehicle
model is. Some labels tell trim package. So, enter all of the labels on the vehicle that
might tell the model.
Step 8. Most times, people pull up to a gas pump so that their vehicle's fuel fill door is on
the same side of the vehicle as the gas pump. If they do that, enter Fueling Side = Near.
Occasionally, people stretch the pump hose across, behind, or under their vehicle because
their fuel fill door is on the far side of their vehicle from the pump. If they do that, enter
Fueling Side = Far.
Step 9. Look at how they put the nozzle in their fuel fill pipe. Enter RSU for
RightSideUp, USD for UpSideDown. Or SW for SideWays.
D-l
-------
High Evaporative Emissions Investigation Field Study
Final Report
Step 10. The typical fueling event ends with the nozzle automatically clicking off and the
customer puts the nozzle back on the pump without further pulling the nozzle handle. If
that happens, enter the Number of Extra Clicks=0. If the customer does not accept the
automatic shut-off, he will attempt to put more gasoline into the tank and each attempt
will produce "extra clicks" as the nozzle automatically turns off again. Count the number
of extra clicks and enter the count on the logsheet.
Sometimes nozzles falsely click off at the beginning of a refueling. Do not count these
clicks.
Step 11. Using your personal timepiece, enter the Nozzle Hang-Up Time (hh:mm:ss)
when the customer puts the nozzle back on the pump.
Step 12. Sometimes during fueling the pump nozzle seems to fail to automatically shut-
off the fuel flow, and there can be a violent gushing of fuel from the vehicle's fill pipe.
This is called a spitback. If a spitback happens for any reason, enter Spitback=Yes,
otherwise enter Spitback=No.
Step 13. Occasionally, a liquid fuel spill can occur during fueling. The spill could be
caused by a spitback, or by repeated clicks as the customer tries to maximize the amount
of fuel in the tank, or by something else. If there is a spill and it comes from the fuel
fillpipe, enter the Spill Source=F. Also, estimate and enter Spill Size as N, T, G, or B.
Step 14. Occasionally, a vehicle will leak gasoline under the vehicle. So, after the vehicle
drives away, look for gasoline puddles under where the vehicle was sitting. Puddles
under vehicles can be from things other than gasoline, such as AJC condensate (water),
green or orange antifreeze coolant, blue windshield washer fluid, engine oil, pink
transmission fluid, brake fluid, or power steering fluid. The smell, color, and viscosity of
gasoline ought to be a giveaway. The other thing is to try to be sure that the gasoline
puddle is from the vehicle that was just there and not an earlier vehicle. If there is a
gasoline puddle under and attributed to the vehicle, enter the Spill Source=U. Also,
estimate and enter Spill Size as N, T, G, or B.
Step 15. Record the action that the station attendant takes in response to spills or
spitbacks by entering N, K, S, and/or C under Attendant Action for none, kitty litter,
spray, or orange cones.
Step 16. Text. If something unusual happens - like spills or spitbacks, please describe
what happened in the full width of the data form lines under the line for the vehicle. Use
complete sentences, so we can tell what happened. Use a lot of words. Don't try to
squeeze a couple of words into the little box provided at the end of the line. For typical
refueling events with no spills or spitbacks, no text description is needed.
Step 17. After the vehicle drives away and after all data entries and any text has been
written, have been made, go back to Steps 2 and 3 and pick the next vehicle to follow.
D-2
-------
High Evaporative Emissions Investigation Field Study
Final Report
Figure D-1. Logsheet for Observations of Randomly Selected Vehicles
Day of Week
Pump Number
License
Make
Model
Fueling Side (Near, Far)
Nozzle
Orientation:
RightSideUp
UpSideDown
SideWays
Number of Extra Clicks
(0=auto shut-off only)
Nozzle
Hang-Up
Time
(hh:mm:ss)
Spitback? (Yes, No)
Spills
Attendant Action
(None, Kitty, Spray, Cone)
Text:
If there
was an
event, tell
about it in
lines
under the
car's data.
State
Plate
(very clear)
Source:
Fill neck
Under car
Size:
Nickel
Tennis
Grapefruit
Bucket
Select next customer who exits car, follow until departure (Look on ground): Record all selected customers whether there is an event or not.
Extra Click = The number of times that the automatic nozzle shut-off is not accepted by the customer.
Spills are caused by the customer clicking the nozzle excessively.
Spitback is a violent gushing of fuel from the fillpipe.
D-3
-------
High Evaporative Emissions Investigation Field Study
Final Report
Appendix E
Estimation of Headspace Vapor Properties for Denver Summer Conditions
-------
High Evaporative Emissions Investigation Field Study
Final Report
The total flow of the artificial releases was 10 gallons/minute, which is a typical fuel
dispensing flow of gas station fuel pumps on their highest pump nozzle setting. According to our
ReddyEvap 2010 calculations (see Table 1), headspace vapor in Denver at summer temperatures
is approximately 50 vol% HC vapor.
We made ReddyEvap 2010 headspace calculations using the following inputs: 8.7 psi
RVP, 10 vol% ethanol, 88 F ambient temperature, 0.83 atm barometric pressure. The partial
pressures were: ethanol 62.53 mmHg, non-ethanol HC 289.96 mmHg. The barometric pressure
was 631 mmHg (=0.83 * 760 mmHg). Therefore, the headspace composition was: ethanol 10
vol%, non-ethanol HC 46 vol%, and air 44 vol%.
The results shown at the bottom of Table E-l show that the estimated headspace vapor
concentration is 4.6 g HC per gallon of headspace vapor.
E-l
-------
High Evaporative Emissions Investigation Field Study Final Report
Table E-1. Reddy Evap 2010 Inputs and Outputs
to Estimate Summer Fuel and Headspace Properties in Denver
**** INPUT DATA ****
Test Fuel RVP
=
8.7psi
Tank Volume
=
18.Ogal
Tank Capacity
=
16.Ogal
Volume Percent Fill
=
50
Tank Orifice Diameter
(inch)=
0.500
Fuel Cap Blow off Pressure =
0.0 inch H20
Tank Pressure Control
Valve =
0.Oinch H20
Atmospheric Pressure
=
0.83atm
Vapor Pressure of 8.7
psi RVP
gasoline at 88.OF =
6.82 psia
Liguid
Vapor
Partial
HeadSpace
Component
MW
Fuel
Pressure
Pressure
Vapor
(mass%)
(mmHg)
(mmHg)
(mass%)
Propane
44.1
0.044
8244.51
7.44
1.411
2M-Propane
58.1
0.472
3102.92
25.42
6.354
Butane
58.1
2.133
2194.44
81. 47
20.364
t-2-Butene
56.1
0.052
2117.36
1.78
0.430
Ethanol
46.1
10.528
83. 43
62.53
12.390
Isobutanol
74.1
0.002
15.82
0. 00
0.000
2M-Butane
72.2
4 .375
849.11
55. 65
17.267
1-Pentene
70.1
0.642
787.39
6. 97
2.102
Pentane
72.2
3.917
639.62
37.58
11.661
2,3-DM-2-Butene
84.2
0.095
163.39
0.21
0.075
2M-2-Butene
70.1
0.789
584.53
6.71
2.023
Cyclopentane
70.1
0.344
401.53
2.03
0.612
2,3-DM-Butane
86.1
2.094
297. 81
8.13
3.011
2M-Pentane
86.2
5 .218
270.92
18.42
6.828
1-Hexene
84.2
0.505
239.23
1.44
0.523
Hexane
86.2
3.048
195.94
7.80
2.889
M-Cyclopentane
84.2
1. 497
178.03
3.49
1.2 62
2, 4-DM-Pentane
100.2
2.056
128.71
3. 02
1.300
Benzene
78.1
2.368
125.30
4 .10
1.378
2,3-DM-Pentane
100.2
2.572
90.94
2 . 67
1.149
3M-Hexane
100.2
3.587
81.73
3.35
1.442
2,2,4-TM-Pentane
114 .2
4.731
65 . 69
3.14
1.541
Heptane
100.2
1.383
61.54
0.97
0.419
M-Cyclohexane
98 .2
1.313
61.76
0.94
0.399
Toluene
92.1
8.363
38.73
3.92
1.554
2,3-DM-Hexane
114 .2
1.738
32 .10
0.56
0.275
3M-Heptane
114 .2
1.317
27.01
0.36
0.175
2,2,5-TM-Hexane
128 .3
1. 420
22 . 95
0.29
0.159
Octane
114.2
0.697
19.59
0.14
0.067
E-Benzene
106.2
2 .230
13. 40
0.31
0.142
m&p-Xylene
106.2
9.390
11.74
1.12
0.511
Nonane
128.3
0.617
6.25
0. 03
0.019
n-Propylbenzene
120.2
2 . 636
4.92
0.12
0.061
1M-3E-Benzene
120.2
3.425
4.29
0.13
0.068
Decane
142.3
7.406
1. 99
0.12
0.071
1,2,3-TM-Benzene
120.2
5 . 658
2.42
0.11
0.059
1,3-DE-Benzene
134.2
1.338
1. 64
0. 02
0.010
Stoichiometric air/fuel ratio of 8.7 psi RVP gasoline vapors at 88.OF =14.45
Air/fuel ratio of 8.7 psi RVP gasoline vapors at 88.OF = 0.35
RVP = 8.7 psi T = 88.0 gHC/gal vapor = 4.639
8.7 psi RVP gasoline vapors at 88.OF contain 55.88 percent HC
Lower flammability of 8.7 psi RVP gasoline vapors at 88.OF = 2.01
Upper flammability of 8.7 psi RVP gasoline vapors at 88.OF =10.12
Molecular Weight of Vapor 66.0
Boiling point of 8.7 psi RVP gasoline at 12.20 psia pressure = 117.4F
E-2
------- |