Citation
Estimating annual bicycle volumes on multi-use paths in Boulder

Material Information

Title:
Estimating annual bicycle volumes on multi-use paths in Boulder
Creator:
Lewin, Amy
Publication Date:
Language:
English
Physical Description:
x, 78 leaves : ; 28 cm

Subjects

Subjects / Keywords:
Bicycle trails -- Colorado -- Boulder ( lcsh )
Trails -- Colorado -- Boulder ( lcsh )
Cycling -- Colorado -- Boulder ( lcsh )
Traffic flow -- Colorado -- Boulder ( lcsh )
Bicycle trails ( fast )
Cycling ( fast )
Traffic flow ( fast )
Trails ( fast )
Colorado -- Boulder ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Bibliography:
Includes bibliographical references (leaves 77-78).
General Note:
Department of Civil Engineering
Statement of Responsibility:
by Amy Lewin.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
|Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
66463260 ( OCLC )
ocm66463260
Classification:
LD1193.E53 2005m L48 ( lcc )

Full Text
*
ESTIMATING ANNUAL BICYCLE VOLUMES ON
MULTI-USE PATHS IN BOULDER
by
Amy Lewin
B.S., Washington University in St. Louis, 1997
A thesis submitted to the
University of Colorado at Denver
in partial fulfillment
of the requirements for the degree of
Master of Science
Civil Engineering
2005


This thesis for the Master of Science
degree by
Amy Lewin
has been approved
by
BruceJanson
/\j/*{
J
Z<3D £T
Date


Lewin, Amy (M.S., Civil Engineering)
Estimating Annual Bicycle Volumes on Multi-Use Paths in Boulder
Thesis directed by Professor Bruce Janson
ABSTRACT
This study provides the first comprehensive analysis of five years of detector data
for two permanent bicycle count stations (representing four locations) on multi-use
paths in Boulder, CO. First, temporal patterns of daily bicycle counts were
explored. A strong linear correlation between high temperature and daily counts
was noted with a slight decrease in counts at temperatures greater than 90 F.
Counts also decreased on days with rain or snow, although this effect was not
linear. In addition, counts decreased on weekends at most locations. The results
of the temporal pattern and weather correlation analyses were used to derive the
factors in the linear regression model that was developed to estimate daily volumes
for those days missing counts. The daily counts and estimates were combined to
generate annual volumes for each year in the study time period. Finally, annual
estimates were adjusted based on model residuals and were used to develop
trendlines for forecasting volumes at these locations in the year 2005. Based on
these estimates, although the volumes fluctuate, three out of the four locations
have exhibited a slightly decreasing trend over the past five years.
1U


This abstract accurately represents the contents of the candidates thesis. 1
recommend its publication.
IV


ACKNOWLEDGEMENT
I would like to thank Bruce Janson, my advisor, as well as Marni Ratzel, Bill Cowern,
Bob Major, Stan Zeedyk, and Randall Rutsch of the City of Boulder. Thanks also to
Joe Racosky and Maggie Boys for their support and advice.


CONTENTS
Figures...................................................................viii
Tables......................................................................x
Chapter
1. Introduction............................................................1
1.1 Purpose of the Study................................................1
1.2 Scope of the Study..................................................2
1.3 Arrangement of the Thesis...........................................7
2. Review of Literature....................................................8
3. Data...................................................................10
3.1 Detectors and Data Collection......................................10
3.2 Data Summary.......................................................12
3.3 Data Issues........................................................13
3.3.1 Missing Data....................................................13
3.3.2 Change in Data Formats.........................................14
3.3.3 Duplicate Data..................................................14
3.3.4 Other Sources of Possible Data Error............................14
4. Methodology............................................................17
5. Results................................................................18
5.1 Summary' Statistics................................................18
5.2 Temporal Patterns..................................................21
5.2.1 By Day of Week.................................................21
5.2.2 By Weekday/Weekend.............................................26
5.2.3 By Month.......................................................30
5.2.4 By Season......................................................34
vi


5.3 Correlation with Weather............................................38
5.3.1 Boulders Climate...............................................38
5.3.2 Temperature.....................................................39
5.3.3 Precipitation...................................................43
5.4 Annual Volume Estimates.............................................49
5.5 Annual Trend Analysis...............................................52
6. Conclusions and Recommendations........................................58
7. Further Research.......................................................62
Appendix....................................................................64
A. Detector Locations and Number of Loops..............................65
B. Detector Images.....................................................66
C. Detector Text File Examples.........................................75
References..................................................................77
vii


FIGURES
Figure
1-1. Boulder Bicycle Facilities and Permanent Count Stations...............3
1-2. ebwbarap Vicinity Map................................................4
1-3. colfthl Vicinity Map.................................................6
3-1. Detector Locations...................................................11
5-1. ebwbarap Daily Count Histogram.......................................19
5-2. colfthl Daily Count Histogram........................................20
5-3. Average Daily Count by Day of Week...................................22
5-4. ebwbarap Daily Counts by Day of Week................................24
5-5. colfthl Daily Counts by Day of Week..................................25
5-6. Average Daily Count by Weekday/Weekend...............................27
5-7. ebwbarap Daily Counts by Weekday/Weekend.............................28
5-8. colfthl Daily Counts by Weekday/Weekend..............................29
5-9. Average Daily Count by Month.........................................31
5-10. ebwbarap Daily Counts by Month......................................32
5-11. colfthl Daily Counts by Month.......................................33
5-12. Average Daily Count by Season.......................................34
5-13. ebwbarap Daily Counts by Season.....................................36
5-14. colfthl Daily Counts by Season......................................37
5-15. Average Daily High Temperature by Month.............................39
5-16. ebwbarap Daily Count by High Temperature............................41
5-17. colfthl Daily Count by High Temperature.............................42
5-18. Average Daily Count by High Temperature.............................43
5-19. ebwbarap Daily Counts by Rainfall...................................44
5-20. colfthl Daily Counts by Rainfall....................................45
viii


5-21. ebwbarap Daily Counts by Snowfall.....................................47
5-22. colfthl Daily Counts by Snowfall......................................48
5-23. ebwbarap Annual Trend Analysis........................................55
5-24. colfthl Annual Trend Analysis.........................................56
5-25. Annual Trend Analysis Summary.........................................57
B-l. Boulder Creek Path-East Side, Looking East.............................66
B-2. Boulder Creek Path-West Side, Looking West.............................67
B-3. End of Broadway Path, Looking North (Towards 13th Street)..............68
B-4. Foothills Path North of cofthl, Looking South..........................69
B-5. Foothills Path, Looking South..........................................69
B-6. Foothills Path, Looking North..........................................70
B-7. Foothills Path at Intersection with Centennial Trail, Looking North....71
B-8. Centennial Trail, Looking East.........................................72
B-9. Colorado Avenue/Foothills Parkway Signal Cabinet.......................73
B-10. colfthl Detector......................................................74
C-l. Comma-Delimited BIN File Example.......................................75
C-2. XML BIN File Example...................................................76


TABLES
Table
3-1. Number of Days with Counts by Month by Year............................12
5-1. Summary Statistics.....................................................18
5-2. Seasonal Factors.......................................................35
5-3. Regression Model Specifications........................................50
5-4. Raw Annual Volume Estimates............................................51
5-5. Adjusted Annual Volume Estimates.......................................53
A-l Detector Locations and Number of Loops..................................65
C-l. Comma-Delimited BIN File Data Dictionary...............................75


1.
Introduction
1.1 Purpose of the Study
In recent years planners and policymakers have paid increasing attention to bicycle
and pedestrian facilities; however, planning for such facilities is often hampered by
inadequate quantitative data. A recent study for the US Bureau of Transportation
Statistics (BTS) stated that existing user count data for such facilities is scarce.
Obtaining such data should be a high priority and would support research, the
planning and design of facilities, as well as policymaking, according to the study
(USDOT 1999).
The City of Boulder, CO presents a rare opportunity. It has collected bicycle count
data at permanent count stations on multi-use paths at various locations since 1998.
However, use of the data collected has been limited to occasional independent
analyses of individual locations over short time periods by City of Boulder staff.
The purpose of the present study is to provide the first comprehensive analysis of
the data, including an examination of bicycle flow and variability characteristics for
two of the permanent count stations over a five-year period. The permanent count
stations collected data continuously, but only a sample of the data was available for
this analysis (as described in Section 3.3.1).
Specific objectives of the study include:
Exploring the temporal patterns of daily bicycle counts, including counts
by day of the week, weekday/weekend, month, season, and correlation
with weather.
1


Developing a method for generating an annual estimate of multi-use path
bicycle users from the sample of data over a five-year period.
Determining the trend of bicycle usage on multi-use paths over those five
years, as well as forecasting future usage.
1.2 Scope of the Study
Boulder offers over 350 miles of dedicated bike facilities, including on-street and
contra-flow bike lanes, designated routes, paved shoulders, soft surface multi-use
paths, and paved multi-use paths (City of Boulder 2005). The focus of this study is
on the paved multi-use path system, which includes over 60 miles of facilities that
serve cyclists, pedestrians, inline skaters, and joggers. The City of Boulder has
permanent count stations at twelve different locations. These locations are shown
in Figure 1-1 and are described in more detail in the Appendix.
The scope of this study includes data for the Years 2000-2004 at the two locations
denoted by the solid stars in the figure, ebwbarap and colfthl. Ebwbarap was
chosen because it was a priority of the City of Boulder and was expected to have
some of the highest volumes in the multi-use path system. Colfthl was chosen
because the data was clean and because its location and adjacent land use implied
that it would be a good contrast for comparison with ebwbarap.
2


Figure 1*1. Boulder Bicycle Facilities and Permanent Count Stations
The ebwbarap station is located at the intersection of the Boulder Creek Path and
the Broadway Path near Arapahoe Avenue and 13th Street in Downtown Boulder,
as shown in Figure 1-2. The adjacent land use is primarily commercial and civic
with some nearby residential uses. The University of Colorado, Boulder Public
Lib ran,', and Boulder High School are all in the vicinity.
The Boulder Creek Path is approximately seven miles long and extends from east
Boulder, near Valmont Reservoir, to the western end of town where Boulder
Canyon and Fourmile Canyon intersect.
3


Figure 1-2. ebwbarap Vicinity Map
The Broadway Path is approximately four miles in length and extends from just
south of Greenbriar Boulevard to the Boulder Creek Path at the ebwbarap count
station. The Broadway Path is adjacent to the University of Colorado (CU) campus
between Baseline Road and University Avenue.
4


The colfthl station is located at the intersection of the Foothills Path and the
Centennial Trail in East Boulder, as shown in Figure 1-3. The adjacent land use is
primarily residential, with an elementary school, a church, and the University of
Colorado Research Park also in the vicinity.
The Foothills Path is approximately three miles long. To the north, the Foothills
Path continues north along Foothills Parkway and connects with an on-street bike
lane on 47th Street and intersects with the Cottonwood Trail. To the south, the
path crosses to the west side of Foothills Parkway (via a bicycle/pedestrian
overpass) and extends south to Thunderbird Lane, where it converts to an on-
street bike lane.
The Centennial Trail is approximately two miles in length and is part of the east-
west bicycle facility network that connects the area near Cherryvale Road with the
University of Colorado Research Park.
5


Figure 1-3. colfthl Vicinity Map
6


1.3 Arrangement of the Thesis
The remainder of the report is arranged in the following manner:
Section 2 provides a review of related literature.
Section 3 is an overview of the detector data, including how data was
collected, a summary of the data collected, and data issues encountered and
mitigated.
Section 4 outlines the overall methodology of the analysis.
Section 5 constitutes the majority" of the document and presents the
results, including summary" statistics, temporal patterns, correlation with
weather, as well as annual volumes estimates and trend analysis.
Section 6 offers conclusions.
Section 7 presents some opportunities for further research.
7


2.
Review of Literature
Given that there is not a lot of bicycle data available in the US, annual estimates of
bicycle usage and trend analyses are fairly rare. As a supplement to the Federal
Highway Administration's National Bicycling and Walking Study, Hunter (1995)
compiled bicycle data from multiple cities that illustrates the inconsistency of bicycle
counting programs1. Where data is available over multiple years no consistent overall
trends are exhibited.
Some cities report results of routine manual count programs that count the same
locations at the same time period each year over several years (e.g., San Luis Obispo,
Gainesville); these counts may support the monitoring of transportation modes, as
defined by the General Plan. Other counts are often limited to project-specific
counts (e.g., the recent 17th Street Bike Lane project in Boulder).
On those occasions in which a fair amount of data exists, several studies have
analyzed temporal patterns (e.g., time of day, day of week, weekday/weekend,
monthly, seasonal) and/or the relationship between weather and cycling. Most of
this research has been performed outside the US.
In one of the few longitudinal analyses of bicycle count data in the US, Niemier
(1996) analyzed AM and PM Peak Period count data at six locations in the Seattle
area over several months in one year. Data was collected manually and cross-
checked with surveys. Niemier found a nonlinear relationship between counts and
high temperature, as well as between counts and rainfall. The author then created a 1
1 For example, the counts included a daily variation of counts in Eugene, OR for one week in 1978
and monthly user volumes at two locations on a recreation trail in Washington, DC for somebut
not allof the months in 1988 and 1989.
8


Poisson model to explain the variability of peak period counts based on location,
month, and high temperature (whether or not it was greater than 50F). Niemeyer
noted significant variability in the PM Peak Period, due to the presence of nonutility
cyclists, and also recommended a functional classification based on user type at each
location.
Richardson (2000) quantified the effects of weather on cycle trips in Australia using
travel survey data, while exploring relationships between utilitarian and recreational
counts and high temperature and rainfall. Richardson observed that recreational
users are more affected by extreme temperatures and rainfall than utilitarian users
are. He developed seasonal factors to predict bicycle volumes based on the weather
factors.
Brandenburg (2004) analyzed data from video monitoring and from on-site
interviews at four locations in Austria to examine the effects of temperature and
precipitation and a thermal comfort index on bicycle counts. As with the
Richardson study, this study confirmed that commuting bicyclists were less weather-
sensitive than recreational bicyclists.
Emmerson (1998) analyzed permanent count station data to try to quantify the
effects of weather on cycle flows at three locations in the UK. He concluded that a
1C increase in high temperature yields a 3% increase in daily volumes. Emmerson's
data suggested that cycle flows are more influenced by maximum temperature than
by rainfall, although both relationships were significant. Flows decreased with any
rainfall at all (i.e., the amount of decrease of a count did not depend on the amount
of rainfall but rather on whether it rained at all).
9


3.
Data
3.1 Detectors and Data Collection
Bicycle count data is collected continuously by 3M Canoga 824T detectors that are
embedded in the concrete of the multi-use paths. Usually the detector hardware
shares a cabinet with adjacent traffic signal controllers, but ebwbarap and the
adjacent nbsbbwy station have their own exclusive cabinet, because the nearest
traffic signal is too far away. Figure 3-1 shows the detector locations at ebwbarap
and colfthl. Each detector can have up to four detector loops associated with it, and
the software allows the user to set up logic-based detection to capture directionality
of counts.
When the detector counts bicycles, it groups the counts into time periods configured
by the user. The City of Boulder has set the time period bins to be two hours in
length. A signal technician manually uploads the data from each detector. During
the upload, the software compiles the data into text files with each row representing
a two-hour count grouped on the even hours (2:01, 4:01, etc.). These rows were
then ultimately aggregated into days for the analysis described in Section 4.
Methodology.
10


(a) ebwbarap
11


3.2 Data Summary
As mentioned in Section 1.1, within the study period (2000-2004) only a subset of
the days have counts. Table 3-1 summarizes the number of days of data available
by month for the ebwbarap and colfthl locations. It is important to note that the
days on which data was collected vary at each location. Shaded cells with italics are
months that include counts on all days.
Table 3-1. Number of Days with Counts by Month by Year
(a) ebwbarap
Year Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec TOTAL %
2000 15 30 31 16 92 20%
2001 19 15 22 31 4 3 31 125 28%
2002 7 13 31 30 4 85 19%
2003 4 12 6 17 13 52 12%
2004 12 3 3 30 31 16 95 21%
ALL 11 12 12 32 55 47 39 49 64 62 35 31 449 100%
(b) colfthl
Year Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec TOTAL %
2000 5 31 23 59 8%
2001 7 28 31 30 15 24 31 4 5 175 24%
2002 7 7 30 31 75 10%
2003 31 12 12 30 8 3 15 3 31 145 20%
2004 31 29 9 24 30 31 31 30 31 16 262 37%
ALL 76 69 52 60 47 33 70 62 39 69 72 67 716 100%
Note: Shaded cells with italics are months that include counts on all days.
The data available represent a good cross-section of month and years, although the
data sample is not as strong for the months of January, February, and March at
ebwbarap. These months have sample sizes less than 30 (a critical number in some
statistical testing), and this will be noted in the monthly analysis. The number of
counts per year at each location will influence the annual estimates; the fewer
counts for that year, the more daily estimates will be used in the annual estimate.
This means, for example, that for the year 2000 at colfthl 8% days of its days will
have counts and 92% will have estimates, while in the year 2004 over one-third
(37%) will have counts, and two-thirds will have estimates.
12


3.3 Data Issues
Before analyzing the data, an important step is to check the quality of the data.
The following sections describe some of the issues encountered with the data and
how those issues were mitigated.
3.3.1 Missing Data
Issue: As mentioned in Section 1.1, although the detectors collect data continuously,
some historical data from the years 2000-2004 was missing due to power outages
and the fact that uploading the data files is currently a manual process.
The detectors have volatile memory, so if a power outage occurs, all data collected
until that point will be lost2. Furthermore, although data is counted automatically,
it must be manually collected. A signal technician for the City of Boulder visits
each detector, hooks up a laptop, and uploads the data from the detector. The data
buffer holds a limited amount of data (approximately 113 days worth of a four-
loop detector3). When the buffer is filled it then writes over the earliest data rows
with the new data. The amount of data collected is thus dependent on the
frequency that staff is available to go out in the field to do the uploads. The City of
Boulder has long-term plans to upgrade this process to allow automatic uploads,
which should increase the amount of data collected.
Mitigation: The sample size was large enough for the majority of the temporal
pattern analysis. The rare occasion on which aggregation of the data yielded a small
2 Newer versions of the detector (e.g. Canoga C924T-EO) do not lose data in the event of a power
failure (3M 2001).
3 As the number of loops decreases, the number of days of data that can be stored in the buffer
increases.
13


sample size (< 30) was noted. To generate annual estimates based on daily
volumes, a process was developed to fill in the missing daily counts with estimates.
3.3.2 Change in Data Formats
Issue: The format of the data files has changed over the years. From the 2000 Nov
files to the 2002July files the data was formatted in a comma-delimited text file.
From 2003 Feb to 2004 Nov the data was in an XML file format.
Mitigation: Common data elements were extracted from both file formats for
loading into the database. Examples of both file formats, as well as a data
dictionary may be found in the Appendix.
3.3.3 Duplicate Data
Issuer. When the data is uploaded, the history is not erased from memory. Thus,
when a data file was uploaded again before the buffer was filled, the new file
contained data that was already stored in a previous data file.
Mitigation: Duplicate data in data files was identified and formatted so that only one
version of the data was loaded into the database.
3.3.4 Other Sources of Possible Data Error
Research has indicated that the electromagnetic detectors are highly accurate (SRF
Consulting Group, Inc. 2003). Furthermore, visual observation of the detectors by
Citv of Boulder staff and the author suggest the detectors accurately count bicycles
that pass the loops. However, there is still the potential for data error based on
14


limitations of the detector configuration and software, as well as human error in the
manual upload process.
3.3.4.1 Error in Counts
Issue: With the current configuration of the detectors, they respond to a
predetermined density of metal (assumed to be a bicycle) that crosses the
electromagnetic field of the loop. If two bicycles cross a loop at the same time, it is
possible that a detector will not detect both bicycles but will instead count one. It
is also possible that a high-metal content stroller would be counted as a bicycle.
Mitigation: It is assumed that these two situations (wThich would offset each other)
do not occur frequendy and that the counts as recorded in the data files, are of
sufficient quality' for use in this analysis.
3.3.4.2 Odd Hour Time Stamps
Issue-. Under the current version of the software when the time changes between
Daylight Saving Time and Standard Time, the software adjusts the time stamp to
correspond to the new time. Because the software normally stamps each time on
the even hour, when the time changes (adding or subtracting an hour) the time
stamp ends up being odd for times after the time change.4
Mitigation: For the purposes of this study the odd hours were adjusted to even
hours so that each day would have 12 time stamps, representing 24 hours.
Although this means that some time stamps were actually on the wrong day, this
4 Changing the software configuration to collect counts per hour would help resolve this problem;
however this would decrease the number of days of data the buffer could hold. Thus, the data
would need to be uploaded more often to prevent data loss.
15


affects only the time-stamps near midnight, which typically represent low counts.
Therefore, this method was judged adequate for this analysis.
3.3.4.3 Incorrect Time Stamps
Issue: The detector does not have a real clock in it5; it has a counter that counts ticks
each 50 milliseconds. The time stamp in the data files is dependent on the time of
the computer uploading the data to convert the ticks into time stamps for the data
files. As with most traffic signal hardware, the detector counter can also drift
slighdy. Sometimes when the technician connected the laptop, he needed to adjust
the laptop time to for the time stamp labels to read *:01.
Mitigation: Even if the time stamps are slightly incorrect, this should not greatly
affect daily counts, because errors would occur in near midnight, when counts are
generally low.
3.3.4.4 Unmatched Duplicate Data
Issue-. Sometimes data files had duplicate data that did not match (possibly a result
of the laptop time adjustment described in Section 3.3.4.3); thus two data files for
the same location had the same date and time stamps but different counts for those
dates and times.
Mitigation: A manual review of the data was undertaken to determine which set of
the data to load.
5 Newer versions of the detectors have real-time clocks, which should improve the quality of the
data (3M 2001).
-Ay.
16


4. Methodology
The detector data text files were provided by the City of Boulder. The data was
cleaned (as described in Section 3.3) and formatted in Microsoft Excel. The Excel
files were then loaded into a Microsoft Access database. Customized queries were
created and run on the data, which was then exported back to Excel. Both Excel
and R (version 2.2.0) were used to perform statistical analyses, including summary
statistics, temporal patterns, and correlation with weather data. A linear regression
model was developed to provide daily volume estimates for those days missing
counts. Annual estimates for each year were then calculated based on both actual
counts and the linear regression model estimates. These estimates were then
adjusted based on the model residual for each year. Linear trendlines were
developed from the raw and adjusted annual estimates for each year. These
trendlines were then used to forecast a range of possible annual bicycle volumes at
the locations studied for the year 2005.
17


5.
Results
5.1 Summary Statistics
As shown in Table 5-1, ebwbarap has 449 full daily counts, or about 25% of all
days during the time period analyzed (2000-2004). Colfthl has 716 full daily counts,
or about 39% of all days during the time period.
Table 5-1. Summary Statistics
Location N* Path Daily Count
Avg Std Dev Std Error Min Max
ebwbarap 449 Boulder Creek Path-East Side 955 444 21 48 2,043
Boulder Creek Path-West Side 868 415 20 40 1,816
colfthl 716 Foothills Path 173 96 4 9 562
Centennial Trail 257 146 5 9 773
*Note: N = sample si%e
At the ebwbarap location, the Boulder Creek Path-East Side average count is
approximately 10% greater than the average count at Boulder Creek Path-West
Side. At the colfthl location, the Centennial Trail average count is almost 50%
greater than the average count on the Foothills Path. In general, ebwbarap daily
counts are three to four times the counts at colfthl; in fact, the maximum counts at
the colfthl locations are even less than the average counts at ebwbarap6. The
relatively high standard deviations for each loop (~50% of the respective average)
indicate a significant amount of variability7 in the daily counts at each location.
This variability is also evident in the histograms of daily count frequency in Figure
5-1 and Figure 5-2.
6 However, it should be noted again that these averages are based on a different number of counts
on different days throughout the study period.
18


Boulder Creek Path-East Side
0 500 1000 1500 2000 2500
Daily Count
(a) Boulder Creek Path-East Side
Boulder Creek Path-West Side
O
O -i
O CO
o
o to
c
3
cr
CD
LL 40 L.
O CM
0 500 1000 1500 2000 2500
Daily Count
(b) Boulder Creek Path-West Side
Figure 5-1. ebwbarap Daily Count Histogram
19


Foothills Path
200
400
Daily Count
600
800
(a) Foothills Path
Centennial Trail
o
o
CM
200
400
Daily Count
600
800
(b) Centennial Trail
Figure 5-2. colfthl Daily Count Histogram
20


The Boulder Creek Path locations have similarly-shaped distributions that are
shifted slighdy to the right, with the East Side having slighdy higher counts overall.
At colfthl, the Foothills Path shows a significant number of dailv counts between
50 and 200, while the Centennial Trail is more evenly distributed over a wider range
of counts.
5.2 Temporal Patterns
The following sections work toward the first objective of this study: to explore the
variability of the data by analyzing various temporal patterns, including day of
week, weekday/weekend, month, and season.
5.2.1 By Day of Week
An analysis of average daily count by day of the week shows no noticeable
variability' at one location, while at the other location, the average daily count varies
noticeably.
Figure 5-3 displays the average daily count by day of week for all four paths. The
highest average daily count is bolded and outlined.
Neither Boulder Creek Path location exhibits significant variation in the average
count by day of the week. For example, on the Boulder Creek Path-East Side the
difference between the highest average daily count (996) and the lowest average
daily count (894) is only 102, or 10%. The average daily count for both Boulder
Creek Path locations is highest on Tuesday and lowest on Friday and Saturday.
21


Average Daily Count by Day of Week
Figure 5-3. Average Daily Count by Day of Week
On the Foothills Path and Centennial Trail, there is a fair amount of variability by
day of the week, mosdy seen in the drop in volumes on Saturday and Sunday. On
the Foothills Path the difference between the highest average daily count (218) and
the lowest average daily count (111) is 107, or 108%. The average daily count for
both the Foothills Path and Centennial Trail is highest on Tuesday and lowest on
Saturday and Sunday.
22


Given that averages alone do not fully describe the data, boxplots were created to
display the distribution of the data. The middle box represents the middle 50% of
the data, with the bold line indicating the median value. The dotted lines, or
whiskers, show the extent of the remaining data and may extend up to 1.5 times the
box length. Any points beyond that are plotted as circles. Thus the boxplots
allows us to check quickly for symmetry (whether the shape looks unbalanced) and
outliers (data points beyond the whiskers) (Verzani 2002).
Although the ebwbarap boxplots in Figure 5-4 show slighdv larger ranges for
Wednesdays at these locations, overall the middle quartiles are fairly similar for all
days. The colfthl boxplots in Figure 5-5 also show the drop in volumes on the
weekends that was evident from the average daily counts, especially at the Foothills
Path location. The outlier on the Foothills Path on 4-Wed is the Bike to Work Day
count from 2004.
23


Boulder Creek Path-East Side
Day of Week
(a) Boulder Creek Path-East Side
Boulder Creek Path-West Side
Day of Week
(b) Boulder Creek Path-West Side
Figure 5-4. ebwbarap Daily Counts by Day of Week
24


Foothills Path
(a) Foothills Path
Centennial Trail
(b) Centennial Trail
Figure 5-5. colfthl Daily Counts by Day of Week
25


5.2.2 By Weekday/Weekend
The data was then analyzed by weekday vs. weekend. As expected at the locations
with no significant variation by day of the week, the variation by weekday/weekend
was minimal, and the location with significant variation by day of the week showed
significant variation by weekday/weekend.
Figure 5-6 displays the average daily count by weekday/weekend for all four paths.
At ebwbarap, average weekend daily counts are slightly higher than average
weekday daily counts for the Boulder Creek Path-East Side, while the opposite is
true for the Boulder Creek Path-West Side.7
At colfthl, average weekend daily counts are significantly lower than average
weekday daily counts for the Foothills Path (-81, or -40%) and slightly lower than
weekday counts for the Centennial Trail (-33, or -12%). This could indicate that
the Foothills Path at this location is used more bv weekday commuters than by
weekend cyclists.8
7 The higher weekend counts at Boulder Creek Path-East Side may be related to its connectivity to
13th Street, which has bike facilities leading to popular Downtown destinations, such as the Farmers
Market (Saturdays from April to October) and the Pearl Street Mall. Those cyclists accessing 13th
Street via the East Side of the Boulder Creek Path do not have to pass the detectors of the West
Side; analysis of the nbsbbwy station data may be able to confirm this travel pattern.
8 The Foothills Path may be perceived as a less attractive north-south route for discretionary" trips
because of its proximity to Foothills Parkway (4 lanes, 45 mph speed limit) and because of the at-
grade intersections at Arapahoe Road, as well as Pearl Parkway, and Valmont Road further to the
north. It will be interesting to see if travel patterns are affected after the planned underpass at
Arapahoe Road is constructed (scheduled for within the next few years).
26


Average Daily Count by Weekday/Weekend
1,200
1.000
600
600
400
200
0
951 J964
Boulder Creek Path-East Side Boulder Creek Path-West Side Foothills Path
Day Type
Centennial Trail
IHWeekday PWeekend |
Figure 5-6. Average Daily Count by Weekday/Weekend
Boxplots of the data by weekday/weekend are shown in Figure 5-7 and Figure
5-8. Although the middle quartiles are fairly similar weekday vs. weekend at each
location, the overall range of the data as denoted by the whiskers is much shorter
on the weekends, especially on the Foothills Path and Centennial Trail.
27


Figure 5-7. ebwbarap Daily Counts by Weekday/Weekend
28


Figure 5-8. colfthl Daily Counts by Weekday/Weekend
29


5.2.3 By Month
Researchers in this field commonly assume that there is a significant amount of
variation in bicycle volumes from month to month, although there is relatively little
data to support this assumption.
An analysis of the data by month confirms that there is indeed a distinct variability
in average daily counts by month, as displayed in Figure 5-9.
At ebwbarap, average daily counts peak in July and August, while February
generally has the lowest average daily counts. The difference between the average
counts for these months is large; at Boulder Creek Path-East Side, Julys average
count (1,385) is over four times the average count in February (267). The volumes
increase steadily, starting in February, although May does not increase at the same
rate as April at the ebwbarap location9. After July and August, the volumes
decrease for the rest of the year.
At colfthl, average daily counts also peak in July and August, while December has
the lowest average daily counts. The Foothills Path peaks in July, while the
Centennial Trail displays consistent average daily counts between June and
September. The difference between the lowest and the highest average daily count
by month is less pronounced at colfthl than at ebwbarap; at Centennial Trail,
Augusts average count (415) is 2.3 times the average count in December (126).
9 This could be related to exams and the end of the semester for students, who are likely to
represent a lot of the bicycle traffic at the ebwbarap location, which is both close to schools (CU
and Naropa University) and to Downtown. There are also many festivals in Civic Center Park that
may block or impede bicycle traffic on the path.
30


Average Daily Count by Month
Figure 5-9. Average Daily Count by Month
Figure 5-10 and Figure 5-11 show boxplots of daily counts by month for each
location. The varying box sizes and length of whiskers indicate a large amount of
variation in the daily counts collected by month. Note that the months with the
smallest range are also those with the lowest number of counts (< 30), so those
results should be interpreted with caution.
31


Boulder Creek Path-East Side
(a) Boulder Creek Path-East Side
Boulder Creek Path-West Side
(b) Boulder Creek Path-West Side
Figure 5-10. ebwbarap Daily Counts by Month
32


Foothills Path
Month
(a) Foothills Path
Centennial Trail
Month
(b) Centennial Trail
Figure 5-11. colfthl Daily Counts by Month
33


5.2.4 By Season
The months were aggregated by season to analyze seasonal trends. For the
purposes of this analysis, spring was defined as March-May, summer was defined as
June-August, fall was defined as September-November, and winter was defined as
December-Februarv.
The average daily count by season is shown in Figure 5-12. At all locations
summer has the highest average daily count, followed by spring and fall and then
winter. At these locations the summer counts were almost 3.5 times as high as the
winter counts. This is in line with the Hunter and Huang study, which noted
summer counts as three to five times as high as winter (Hunter 1995).
34


Seasonal factors were developed by normalizing the seasonal average daily counts
to the summer average daily count. Thus, the seasonal factors are the ratios of the
average daily count of each season to the average daily count of summer, the
highest season, as shown in Table 5-2. Suggested rules of thumb for planning are
also given, based on the analysis of the four locations.
Table 5-2. Seasonal Factors
Season Average Daily Count as a Percent of Summer
Boulder Creek Path- East Side Boulder Creek Path- West Side Foothills Path Centennial Trail Suggested Rule of Thumb
Spring1 69% 72% 67% 64% 70%
Summer2 100% 100% 100% 100% 100%
Fall3 65% 69% 61% 62% 65%
Winter4 22% 23% 37% 32% 30%
Notes:
1. Spring: March-May.
2. Summer: June-August.
3. Fall: September-November.
4. Winter: December-Feburary.
In general spring is slighdy higher than fall. At ebwbarap spring and fall are
typically 65-72% of the summer counts, while winter is only 22-23% of summer.
This could be because the fine summer weather draws recreational and commuter
cyclists to this very popular and scenic path. At colfthl spring and fall are 61-67%
of the summer counts, while winter is 32-37% of summer.
Figure 5-13 and Figure 5-14 show boxplots of daily counts by season for each
location. Most of the daily summer counts are greater than the majority of all other
counts. As expected, the distributions of spring and fall look similar, and winter
has a much smaller range of data than the other seasons.
35


Boulder Creek Path-East Side
(a) Boulder Creek Path-East Side
Boulder Creek Path-West Side
(b) Boulder Creek Path-West Side
Figure 5-13. ebwbarap Daily Counts by Season
36


Foothills Path
(a) Foothills Path
Centennial Trail
(b) Centennial Trail
Figure 5-14. colfthl Daily Counts by Season
37


5.3 Correlation with Weather
As shown in Sections 5.2.3 and 5.2.4, daily counts exhibit a significant amount of
monthly and seasonal variation. Temperature and precipitation are two important
weather characteristics that influence bicycle volumes (Niemeier 1996; Richardson
2000), so these were tested for correlation with the bicycle counts. Daily weather
data, including high temperature, low temperature, rainfall, snowfall, and snow
depth was obtained from the National Oceanic & Atmospheric Administration
(NOAA) Climate Diagnostic Center.10
5.3.1 Boulders Climate
The average high temperature in Boulder is 65 F, and the average annual rainfall is
19. Furthermore, Boulder typically receives nearly 82 of snow each year
(Western Regional Climate Center 2005). It should be noted that the weather is
mercurial. As quoted in a recent guide book:
Foot-deep snow and below-zero temperatures can suddenly
transform to breezy 60-degree, T-shirt weather by Boulders
capricious Chinook snow-eater winds. .The dreaded high winds,
Boulders biggest weather affliction, can come anytime but are
worst in winter and can continue for days. It can be so windy that
schools wont let voung children leave the building without an
adult. (Schlender 1995)
Because of the prevalence for high winds in Boulder, it was expected to be another
determining factor for bicycle volumes. Unfortunately, daily wind data was
surprisingly difficult to find and was ultimately unavailable for this analysis.
10 Accessed at http://www.cdc.noaa.gov/Boulder/data.daily.html
38


5.3.2 Temperature
As stated by Richardson (2000), changes in temperature are indicative of relatively
predictable seasonal climate variations. Average temperature per month is
presented in Figure 5-15. As shown in the figure, although there are occasional
deviations, the overall curve for the five years has remained a fairly consistent bell-
shape, with the peak in July.
Average Daily High Temperature by Month
-2000
--2001
2002
2003
*2004
Average
Figure 5-15. Average Daily High Temperature by Month
39


Figure 5-16 and Figure 5-17 show scatterplots of daily count by high temperature.
The linear trendline in each plot shows a good fit (with R2 ranging from 0.60 to
0.82). These values are then aggregated in Figure 5-18, which shows the average
daily count by high temperature. The data covers virtually all temperatures
between 20 F and 101 F. At all four locations, the function generally increases
(i.e., the daily counts increase as the high temperature increases) with a slight drop
off in counts after about 90 F, especially at the ebwbarap locations. Overall, daily
high temperature appears to have a strong linear relationship with daily bicycle
counts.
40


Boulder Creek Path-East Side
Daily Count by High Temperature
Trend^ne
y=22.$2S>-840.n
R' 0.9239
a East Side
----Linear Regression
(a) Boulder Creek Path-East Side
Boulder Creek Path-West Side
Daily Count by High Temperature
Trendline
y = 20.434* 559.96
R2 = 0.7542
A West Side j
----Linear Regression |
(b) Boulder Creek Path-West Side
Figure 5-16. ebwbarap Daily Count by High Temperature


Foothills Path
Daily Count by High Temperature
1,000 T
800 +
I
0
0
Trendline
y = 4 i486x 84.38
R' = 0.6037
Foothills Path
----Linear Regression
(a) Foothills Path
Centennial Trail
Hi Temp vs. Daily Count
Trendline
y = 7.1729* 187.48
Ri = 0.7848
| A Centennial Trail
|----Linear Regression
(b) Centennial Trail
Figure 5-17. colfthl Daily Count by High Temperature
42


Average Daily Count by High Temperature
i Boulder Creek Path-East Side Boulder Creek Path-West Side X Centennial Trail A. Foothills Pathl
Figure 5-18. Average Daily Count by High Temperature
5.3.3 Precipitation
Precipitation, such as rainfall and snowfall represents smaller-scale variabilitythe
day-to-day, short-term changes in weather conditions. It was surmised that
precipitation could also be an influencing factor in bicycle volumes. Note that in
Boulder it is somewhat rare for precipitation to occur throughout the whole day.
5.3.3.1 Rain
Figure 5-19 and Figure 5-20 show scatterplots of daily counts by rainfall. The
plots are inconclusive, although intuitively, increasing daily rainfall should yield
lower daily counts. Given the results of previous research, a binary flag/indicator
of rainfall was proposed to be tested in a regression model.
43


Boulder Creek Path-East Side
Daily Count by Rainfall
2,500

>
1. *1
"
T %
0 0.2 0.4 0.6 0.8 1
Rainfall (inches)
(a) Boulder Creek Path-East Side
Boulder Creek Path-West Side
Daily Count by Rainfall
(b) Boulder Creek Path-West Side
Figure 5-19. ebwbarap Daily Counts by Rainfall
44


Foothills Path
Daily Count by Rainfall
Rainfall (inches)
(a) Foothills Path
Centennial Trail
Daily Count by Rainfall
Rainfall (inches)
(b) Centennial Trail
Figure 5-20. colfthl Daily Counts by Rainfall
45


5.3.3.2 Snow
Figure 5-21 and Figure 5-22 show scatterplots of daily counts by snowfall. As
with the rainfall plots, the patterns are inconclusive, although intuitively, increasing
daily snowfall should yield lower daily counts. Again, given the results of previous
research, a binary flag/indicator of snowfall was proposed to be tested in a
regression model.
46


Boulder Creek Path-East Side
Daily Count by Snowfall
Snowfall (inches)
(a) Boulder Creek Path-East Side
Boulder Creek Path-West Side
Daily Count by Snowfall
0 1 2 3 4 5 6
Snowfall (inches)
(b) Boulder Creek Path-West Side
Figure 5-21. ebwbarap Daily Counts by Snowfall
47


Foothills Path
Daily Count by Snowfall
Snowfall (inches)
(a) Foothills Path
Centennial Trail
Daily Count by Snowfall
Snowfall (inches)
(b) Centennial Trail
Figure 5-22. colfthl Daily Counts by Snowfall
48


5.4 Annual Volume Estimates
This section addresses the objective of developing a method for generating an
annual estimate of bicycle usage. Because daily counts for ever)7 day were not
available during the time period studied, volumes for those days that dont have
counts had to be estimated. Then the daily counts and estimates were combined
for comparison by year.
To generate daily estimates, the model was developed based on the key
observations from the analysis in Sections 5.2 and 5.3. These key observations
included:
There is a strong linear relationship between high temperature and daily
counts
There is a slight decrease in counts at temperatures greater than 90 F
Counts may decrease on days with rain or snow
Counts decrease on weekends at most locations.
Separate specifications were developed for each path. Although this approach
increases the complexity of the process, it was expected to also increase accuracy
and would highlight the particular factors that are significant at each location.
The resulting model specifications are shown in Table 5-3. HiTemp and Over90
were the only factors that were statistically significant for all locations. As
expected, the HiTemp coefficient is positive, because as the daily high temperature
increases, the daily count increases; however, all other variable coefficients are
negative, including the intercept. The adjusted R2 values range from 0.78 to 0.85
and indicate that overall each model does a good job of explaining the variation in
the counts.
49


Table 5-3. Regression Model Specifications
Factor ebwbarap colfthl
Boulder Creek Path-East Side Boulder Creek Path-West Side Foothills Path Centennial Trail
Coeff. t-stat Coeff. t-stat Coeff. t-stat Coeff. t-stat
Intercept -674.53 -17.37 -572.06 -13.29 -75.31 -11.76 -184.98 -18.38
HiTemp1 23.96 43.87 21.55 35.76 4.44 44.18 7.51 48.69
RainFlag2 -109.13 -5.26 -131.19 -5.73 - - -29.66 -4.75
SnowFlag3 -85.60 -2.24 -91.02 -2.16 - - -23.74 -2.83
WeekendFlag4 - - -60.20 -3.06 -85.98 -23.07 -41.64 -8.19
Over905 -45.94 -7.26 -44.84 -6.42 -11.55 -6.02 -20.90 -7.89

Adjusted R2 0.85 0.79 0.78 0.82
Standard Error 171.33 188.95 45.14 61.52
Comments Weekend not significant Rain, snow not significant
Notes:
1. HiTemp: high temperature of the day.
2. RainFlag: binary flag (0 or 1) indicating whether or not it rained that day.
3. SnowFlag: binary flag (0 or 1) indicating whether or not it snowed that day.
4. WeekendFlag: binary' flag (0 or 1) indicating whether or not it the day was a weekend.
5. Over90: number of degrees over 90 of HiTemp.
As expected based on the analysis in Section 5.2.2, the weekend indicator flag was
not significant at Boulder Creek Path-East Side. At Foothills Path the rain and
snow flags were not significant. This would indicate that the Foothills Path is
largely a commuter corridor for cyclists.11 Previous research by Nankervis,
Richardson, and Brandeburg has shown that commuters are not as affected by the
presence of precipitation as are other cyclists.
Before using the model to generate daily estimates, an adjustment was made to
prevent the type of unreasonable estimates that often occur with a linear model at
data extremes. This model relies heavily on the high temperature. Although the
11 Time of day analysis could also help confirm this supposition of Foothills Path.
50


linear relationship is quite strong overall, for days with vert low temperatures (e.g.,
< 30 F), it is possible for the model to predict a negative bicycle volume. Given
that a negative bicycle volume is logically impossible, and the data contained
absolutely no daily counts of 0, the minimum count for each location (ranging from
9 at Foothills and Centennial to 48 at Boulder Creek Path-East Side) was
substituted when a negative volume was predicted.12
The model was thus applied with the minimum value substitutions as described
above. The raw yearly totals are shown in Table 5-4, with all values rounded to
the nearest hundred. Based on these estimates, the highest yearly volume was in
2002 for both locations on the Boulder Creek Path, 2001 for the Foothills Path and
2004 for the Centennial Trail. Overall, the range of volumes has not changed
significantly; the difference between the highest and lowest volumes is less than
10%.
Table 5-4. Raw Annual Volume Estimates
Year ebwbarap colfthl
Boulder Creek Path-East Side Boulder Creek Path-West Side Foothills Path Centennial Trail
2000 308,700 283,200 65,200 99,100
2001 306,800 283,700 68,800 98,400
2002 315,900 285,700 67,100 101,700
2003 302,400 266,700 64,400 98,600
2004 289,600 265,000 65,800 103,400
TOTAL 1,523,400 1,384,300 331,300 501,200
Average 304,700 276,900 66,300 100,200
12 Although negative predictions occurred quite infrequently (less than 4% of all ebwbarap and
colfthl estimates) this is, admittedly, an imperfect methodology and is likely to bias estimates at low
temperatures. A different model structure (e.g., logarithmic) could prevent negative estimates.
51


5.5 Annual Trend Analysis
This section addresses the analysis of annual trends, which is the final objective of
this study. Before analyzing annual trends, another limitation of the model should
be noted: the model uses the same factors and coefficients for all five years, which
implies that the factors influencing travel behavior have not changed over time.
Although the models presented here still show a relatively good fit over all the data,
it is quite possible that external factors (e.g., population, employment, gas prices,
network connectivity7) influencing travel behavior have indeed changed over time.
The model output was thus adjusted to minimize possible bias.
To check for possible bias by year, the average residual13 for each year was
calculated. A positive residual indicated that the model was underpredicting the
counts for that year and location, while a negative residual indicated that the model
was overpredicting the count for that year and location. The average residual was
then divided by the average count to determine the adjustment factor for that year
and location. These adjustment factors were then applied to their respective
estimates (note that the adjustment factors were not applied to the actual counts,
only to estimates). The original and adjusted annual estimates provide a reasonable
range for comparison among years and are shown in Table 5-5. The table also
shows the adjustment factor applied to the estimates, as well as whether the model
was overpredicting or underpredicting the counts for that year. It is expected that
the actual annual volume lies somewhere between the original and the adjusted
counts.
13 Residual = Actual Count Model Predicted Count
52


Table 5-5. Adjusted Annual Volume Estimates
(a) ebwbarap
Boulder Creek Path-East Side Boulder Creek Path-West Side
Year Original Adjusted Adj. Factor1 Over / Under2 Original Adjusted Adj. Factor1 Over/ Under2
2000 308,700 319,200 +4.7% Under 283,200 299,200 +7.9% Under
2001 306,800 313,300 +3.5% Under 283,700 296,200 +7.5% Under
2002 315,900 318,100 +1.0% Under 285,700 283,700 -1.0% Over
2003 302,400 283,000 -7.6% Over 266,700 188,100 -33.7% Over
2004 289,600 274,600 -7.2% Over 265,000 257,100 -4.2% Over
(b) colfthl
Foothills Path Centennial Trail
Year Original Adjusted Adj. Factor1 Over / Under2 Original Adjusted Adj. Factor1 Over/ Under2
2000 65,200 66,500 +2.2% Under 99,100 101,100 +2.4% Under
2001 68,800 71,700 +8.5% Under 98,400 97,300 -2.0% Over
2002 67,100 62,800 -7.2% Over 101,700 90,800 -12.1% Over
2003 64,400 59,700 -10.9% Over 98,600 93,400 -7.7% Over
2004 65,800 65,700 -0.7% Over 103,400 104,500 +5.4% Under
Notes:
1. Adj. Factor: adjustment factor that is applied to the estimates only; therefore, Adjusted does
*not* equal Original (1 + Adj.Factor).
2. Over/Under: overpredicted or underpredicted; indicates how model performed for that year.
In general, the model underpredicted values for the year 2000 and overpredicted
for the year 2003. The remaining years varied by location. Most of the adjustment
factors were less than 10%, with the exception of Boulder Creek Path-West Side
and Foothills Path in 2003, and Centennial Trail in 2002. Initial assessment
indicates that the lower than average 2003 counts on the Boulder Creek Path-West
Side probably occurred because of the Broadway Reconstruction Project, which
often closed the Boulder Creek Path just west of the West Side count station. The
other colfthl locations might also have had construction-related decreases in
volume, but this has not been confirmed.
53


The values in the above table were plotted and trendlines were developed based on
the raw and adjusted volumes. The trendlines were then used to forecast raw and
adjusted estimates for the year 2005. The final forecast for 2005 is given as the
range between these two estimates, as shown in Figure 5-23 and Figure 5-24.
Figure 5-23 shows the trendline with and without the 2003 estimate at the Boulder
Creek Path-West Side. The object of the trendline is to show a general pattern of
behavior. The year 2003 at this location appeared to have atypical counts because
of extenuating circumstances (i.e., extensive construction in the vicinity) and
influences the trendline to a great extent, possibly leading to an underprediction of
2005. It thus made sense to base the 2005 forecast for this location on the
trendline without the 2003 estimate. When 2005 detector data is available it is
recommended to apply the model to that data to test ranges forecasted here and to
possibly refine the model.
Based on this methodology7, the overall trend over the past five years at the Boulder
Creek Path at ebwbarap and the Foothills Path at colfthl is a slight decrease in
annual volumes. However, the annual volumes do fluctuate, as shown in Figure
5-25, which displays the adjusted annual volume estimates and percent change
from year to year. As expected, the biggest change occurred at Boulder Creek
Path-West Side between the years 2002 and 2003. In 2004 this location recovered,
with a 37% increase. Boulder Creek Path-East also experienced a decline between
the years 2002 and 2003, possibly also related to the lack of continuity of the path
due to the construction. It is interesting to note that the volumes move
independendy at all locations; between each set of years there is always at least one
location that increases when the others decrease (e.g., Foothills Path from 2000-
2001) or decreases when the others increase (e.g., Boulder Creek Path-West Side
from 2001-2002).
54


Boulder Creek Path-West Side
Annual Trend Analysis
(b) Boulder Creek Path-West Side
Figure 5-23. ebwbarap Annual Trend Analysis
55


Centennial Trail
Annual Trend Analysis
100,000
90,000
IU 70,000
60,000
50,000
2000 2001
2002 2003
Year
3
2004 2005 Est.
_ 2005 Forecast =
'98,000-102,000
Raw
Adjusted
-Linear (Raw)
- Linear (Adjusted)
(b) Centennial Trail
Figure 5-24. colfthl Annual Trend Analysis
56


Annual Trend Analysis
Estimated Annual Volume by Year
(Adjusted)
Figure 5-25. Annual Trend Analysis Summary
57


6.
Conclusions and Recommendations
The first objective of this study was to explore the temporal patterns of daily
bicycle counts, including counts by day of the week, weekday/weekend, month,
season, and weather. Key observations include:
Boulder Creek Path-East Side and -West Side had average counts that were
three to four times higher than the Foothills Path and Centennial Trail
counts.
At all locations, Tuesdays had the highest counts by day of the week. This
would probably be the best day to perform spot counts at other locations
to capture peak daily volumes.
At all locations except Boulder Creek Path-East Side, weekends generally
had lower counts than weekdays.
At all locations, July and August had the highest average daily counts, while
February and December had the lowest average daily counts. This is
somewhat surprising, considering that it is often assumed that students
comprise the majority of cyclists; however, most students are not in
Boulder during the summer. Therefore, the cycling population in Boulder
undoubtedly includes both students and the general population.
Summer typically had the highest average daily counts, and winter the
lowest.
Average spring and fall counts were approximately 65-70% of average
summer counts.
Average winter counts were approximately 30% of average summer counts.
The average daily high temperature increased steadily from February7 and
peaked in J uly.
58


As the average daily high temperature increased, bicycle counts increased
linearly; however, high temperatures over 90F typically showed a slight
decrease in counts.
Scatterplots of counts on days with rain are inconclusive, although
intuitively, increasing daily rainfall should yield lower daily counts.
Similarly, scatterplots of counts on days with snow are inconclusive,
although intuitively, increasing daily snowfall should yield lower daily
counts.
The second objective of the study was to develop a method for generating an
annual estimate of multi-use path bicycle users from the sample of data over a five-
year period. Key observations include:
It is possible to create linear regression models with weather and temporal
factors that explain the variability in counts.
The model specifications here generally explained ~80% of the variation in
counts and provided a relatively simple method for estimating annual usage
on multi-use paths in Boulder.
The final objective of the study was to determine the trend of bicycle usage on
multi-use paths over the past 5 years and forecasting future usage. Key
observations include:
The model underpredicts volumes for some years and overpredicts others,
so adjustment factors based on residual bias for each year and location were
applied.
Based on this methodology7, annual volumes at the Boulder Creek Path at
ebwbarap and the Foothills Path at colfthl show a slight decreasing trend in
volumes over the past five years. The volumes in recent years may7 be lower
because the past two years have seen an increase in days with precipitation,
59


which would cause the model to yield lower bicycle volumes for a given
temperature. It should be noted that these are only two sets of detectors
and do not necessarily imply that bicycle volumes in Boulder are decreasing
in general. When 2005 detector data is available it is recommended to apply
the model to that data to test these forecasted ranges.
It is unlikely to know from the data at these two locations whether bicycle
use in general is increasing or decreasing.
Although Boulder has more bicycle count data than most municipalities in the US,
the software and data collection issues (see Section 3.3) have hindered data
collection. Recommendations to improve data quality7 include:
Collect data more frequently. Detectors with four loops can only store
up to 113 days (just under four months) of data. Collecting data every7 two
months should assure that no data is overwritten and should also minimize
the drift of the detectors. If resources are too limited, one or more high
priority7 detectors could be chosen from which data is uploaded more
frequendy than other locations. This approach would assure large data
samples at least a subset of locations.
Upgrade to automated collection of data. Upgrading to automated
collection of data has been on the Citys wish list for several years. Once
this is set up, it should require minimal maintenance, freeing up City7
resources.
Upgrade to 3M Canoga 900 series detectors. The 900 series detectors
have real-time clocks and non-volatile memory7. These features would
improve accuracy of the time-stamps on the data while preventing data loss
because of power outages.
60


It is also recommended to add permanent count stations to new multi-use path
facilities and to existing paths when they are dug up for improvements. Collecting
and analyzing data from even more locations would provide a more comprehensive
picture of bicycle travel patterns and characteristics on multi-use paths in Boulder.
61


7.
Further Research
The results of the model presented here are promising but offer the opportunity
for refinement. It would be helpful to test for the significance of a factor based on
wind, if the data is made available. Adding other socioeconomic factors not
analyzed in this study may also prove helpful. Other research has also incorporated
holidays as a factor influencing bicycle counts (Emmerson 1998).
The volume adjustment process could also benefit from refinement. Although it is
year-specific, it is still a generalized approach to account for external factors, and
other approaches could be tested.
The analysis presented here was based on two locations only, so it may be
premature to generalize to other locations in Boulder or to other cities. It would be
worthwhile to analyze the ten other sets of detectors to test and refine the models
presented here. It might be possible to look for patterns in the temporal analysis
on which to develop a classification scheme. Time of day analysis could also
support this effort. The locations could then be analyzed as a system and along
corridors (e.g., Arapahoe, Foothills, Broadway). In particular, Foothills corridor
counts should be helpful, because there are not many entry' and exit points along
the path between detectors. It is also recommended to confirm travel patterns at
the ebwbarap detectors by analysing nbsbbwy. Such analyses would yield a better
understanding of travel behavior and help with planning facilities.
62


Furthermore, it may be useful to cross-check these results with traffic counts and
with travel surveys, such as the Downtown Travel Survey, or the annual travel
diaries that are collected. These efforts would work towards the goal of supporting
a comprehensive database related to bicycle and pedestrian facilities.
63


APPENDIX
A. Detector Locations and Number of Loops
B. Detector Images
C. Detector Text File Examples
64


A. Detector Locations and Number of Loops
Table A-l Detector Locations and Number of Loops
M jUliflwk Loop 2 m .. loops tm Loop 4 ffltt
1 arap38th Arapahoe Path @ 38th into 18 Arapahoe Path (WB) Arapahoe Path (EB)
2 bwybln Broadway Path @ Baseline int002 Broadway Path (None)
3 bwytmesa Broadway Path @ Table Mesa int043 Southwest Comer (SB) Southwest Corner (NB)
4 colfthl Foothills Path @ Centennial and Centennial Path @ Foothills int097 Foothills (NB) Foothills (SB) Centennial (EB) Centennial (WB) j
5 crkebwb Boulder Creek Path @ 4000 Arapahoe Arap4000 Creek Path East (EB) Creek Path East (WB) Creek Path West (EB) Creek Path West (WB)
6 ebwbarap Bldr Crk Path @ 13th Arap13th East Side (EB) East Side (WB) West Side (EB) West Side (WB)
7 nbsbbwy Broadway Path @ 13th Arap13th North Side (NB) North Side (SB) South Side (None)
8 nenw Arapahoe Path @ Foothills and Foothills Path @ Arapahoe into 19 Foothills Path (None) Arapahoe Path (None)
9 nese Foothills Path @ Peart Parkway int090 FHills Path NE (NB) FHills Path NE (SB) FHills Path SE (None)
10 nwse Pearl Parkway Path @ Foothills int090 Pearl Path NW (EB) Pearl Path NW (WB) Pearl Path SE (None) Pearl Path SW (None)
North, Near South, Near
11 prl55th 55th Path @ Pearl Parkway int014 Pearl Underpass (None) Pearl Underpass (None)
12 rspkarap Research Park Path @ 4000 Arapahoe Arap4000 To Research Park (NB) To Research Park (SB) From Arap. Rd. (WB) To Arap. Rd. (EB)


B. Detector Images
Figure B-1. Boulder Creek Path-East Side, Looking East
Notes: Boulder High School field is on the left, Boulder Creek is on the right behind
the trees.
66


Notes: The cyclist is passing the West Side detector loops and is approaching the
underpass under Arapahoe Avenue. The path then continues under Broadway
towards the Boulder Public Library.
67


Figure B-3. End of Broadway Path, Looking North (Towards 13th Street)
Notes: This is an at-grade intersection with Arapahoe Avenue. It has a striped
pedestrian crossing where cars are required to stop if a pedestrian/cyclist is present.
The signal cabinet for ebwbarap and nbsbbwy is on the right. Detector loops are in
quadrupole formadon in concrete._______________________________________________
68


Figure B-4. Foothills Path North of colfthl, Looking South
bfotes-FoothiUsI^^ the right, the bicycle/pedestrian overpass is in view.
Figure B-5. Foothills Path, Looking South
Notes: The cyclist is crossing the detector loops. The bicycle/pedestrian bridge is
accessed on the right on a relatively steep and curving grade._____________________
69


Figure B-6. Foothills Path, Looking North
Notes: The detector loops are in the typical quadrupole configuration.
70


Figure B-7. Foothills Path at Intersection with Centennial Trail, Looking North
Notes: The Centennial Trail connects from the right. Boulder Bikeway signs indicate
directions to the University7 of Colorado and to Arapahoe Avenue._______________________
71


Figure B-8. Centennial Trail, Looking East
Notes: The cyclist is about to cross the Centennial Trail detector loops.
72


Figure B-9. Colorado Avenue/Foothills Parkway Signal Cabinet
Notes: colfthl detector is located in this cabinet. The technician connects the laptop
to the detector to upload the detector data.________________________________________
73


74


C. Detector Text File Examples
C800IS Binned Data File VI.0
1,20: 01,07/03/01,28,0.8%,22,0.8%,21,0.4%,18,0.4%,
2,22: 01,07/03/01,24,0.4%,6,0.4%,14,0.0%,1,0.0%,
3,00:01,07/04/01,6,0.0%,0,0.0%,6,0.0%,1,0.0%,
4,02:01,07/04/01,5,0.0%,1,0.0%,3,0.0%,1,0.0%,
5,04:01,07/04/01,4,0.0%,12,0.0%,5,0.0%, 10,0.0%,
6,06:01,07/04/01,66,1.6%,81,1.6%,60,1.6%, 73,2.0%,
7,08:01,07/04/01,108,4.3%,160,4.3%,85,4.3%,168,4.7%,
8,10:01,07/04/01,171,5.5%,178,5.1%,142,4.7%,167,5.5%,
9,12:01,07/04/01,144,4.7%,151,4.7%,119,4.3%,137,5.1%,
10,14:01,07/04/01,111,3.5%,115,3.1%,89,3.5%,91,3.5%,
Figure C-1. Comma-Delimited BIN File Example
Table C-1. Comma-Delimited BIN File Data Dictionary
Column Field Example
1 Line number 1
2 Start Time of 2-Hour Bin 20:01
3 Date 07/03/01
4 Loop 1 Count 28
5 Loop 1 Occupancy 0.8%
6 Loop 2 Count 22
7 Loop 2 Occupancy 0.8%
8 Loop 3 Count 21
9 Loop 3 Occupancy 0.4%
10 Loop 4 Count 18
11 Loop 4 Occupancy 0.4%
75


Figure C-2. XML BIN File Example

;
r
{BA4AFC9D-D6Dl-4200-8A0F-
18A0F64C19F7}

< Created By > Sta n Zeedy k

5/9/2003 12:35:11 PM
1.0

2
C824-F
0060659BG4FN
C800 Vl.l
255


r
r
09/28/02
00:01:48
2
0
l
0
-l

-l
100
______
____________________________________
76


REFERENCES
3M Intelligent Transportation Systems. 1997. Interface software user guide: Canoga vehicle
detection system ITS link C800IS inteface and data acquisition software. St. Paul,
MN: 3M. p 69-72.
3M Intelligent Transportation Systems. 2001. Installation instructions: Canoga vehicle
detection system model C924TE-0 vehicle detector. St. Paul, MN: 3M. p 10.
Brandenburg C, Matzarakis A, Arnberger A. 2004. The effects of weather on
frequencies of use by commuting and recreation bicyclists. Berichte des
Meteorologischen Institutes der Universitdt Freiburg 12: 189-197.
City of Boulder. 2005.
.
Accessed 2005 Oct 18.
Emmerson P, Rvley T, Davies D. 1998. The influence of weather on cycle flow.
Traffic Eng Control 39: 238-243.
Hunter W, Huang, H. 1995. User counts on bicycle lanes and multiuse trails in the
United States. Trans Res Rec 1502: 45-57.
Nankervis M. 1999. The effect of weather and climate on bicycle commuting. Trans
Res Ft A 33: 417-431.
Niemeier D. 1996. Longitudinal analysis of bicycle count variability: results and
modeling implications. J ofTransp Eng, 122(3): 200-206.
Richardson A. 2000. Seasonal and weather impacts on urban cycling trips. jTUTI]
The Urban Transport Institute, Report 1-2000. Victoria, Aus: TUTI.
Schlender S, Glenn R. 1995. The insiders guide to Boulder & Rjocky Mountain National
Park. 2nd ed. Boulder, CO: Boulder Publishing Co., Inc. p 8.
SRF Consulting Group, Inc. 2003. Bicycle and pedestrian detection final report.
Minneapolis, MN: Author.
U.S. Department of Transportation, Bureau of Transportation Statistics. 2000.
Biycle and pedestrian data: Sources, needs, and gaps. (BTS00-02). Washington, DC:
Author.
77


VerzaniJ. 2002. simpleRUsing Rfor introductory statistics. project.org/doc/contrib/Verzani-SimpleR.pdf>. Accessed 2005 Oct 18.
Western Regional Climate Center. 2005 Jul 22. < http://www.wrcc.dri.edu/cgi-
bin/cliMA!N.pl?coboul>. Accessed 2005 Oct 18.
78