The 3rd International Conference on Transportation Information and Safety, June 25 – June 28, 2015, Wuhan, P. R.

China

Annual Average Daily Traffic Estimation from

Short Traffic Counts
LUO Zongfan ZHONG Ming
British Columbia Ministry of Transportation and Intelligent Transport System Research Center
Infrastructure Wuhan University of Technology, China/National
Prince George, Canada Engineering Research Center for Water Transport
Safety
Wuhan, China

Abstract—Highway agencies have gathered traffic data for a Currently, the traditional factor approach as a convention
wide range of engineering and management purposes since the prediction method is being widely used for AADT estimation
early days. Traffic summary statistics, such as annual average from short period traffic counts by highway agencies. In recent
daily traffic (AADT) and design hourly volume (DHV), estimated year, neural networks have received considerable attention as
from such data are the foundation of decision-makings related to more sophisticated alternatives to the conventional prediction
the planning, design, operation and management of highway methods. There are various claims about the usefulness of
transportation system. Traditional factoring method is neural networks for predictions based on insufficient data
recommended by the Federal Highway Administration (FHWA) (Sharma 1999) [2]. Some researchers have also conducted
Traffic Monitoring Guide (TMG 2001) [4] for estimating AADTs
studies on using regression approach for AADT estimation.
from short-period or sample traffic counts. Several studies in the
past have attempted to rationalize duration and schedules of This study employed data from permanent traffic counts
sample traffic counts that are shorter than 24 hours in length. (ATRs) and focused on reliability of AADT estimates from 1-
This study focuses on reliability of AADT estimates from 1-day, day, 2-day, 3-day and 5-day traffic counts.
2-day, 3-day and 5-day traffic counts. The results of this study
reveal that shorter counts (1 or 2-day counts) conducted between
Monday and Wednesday) may produce AADT estimates as good II. LITERATURE REVIEW
as the estimates from longer counts (5-day counts). The analysis A commonly used form of mathematical relationship for
and findings of this study on the adequacy of short period traffic estimating AADT from sample counts of shorter duration than
counts from the perspective of AADT estimation will help 24 hours is that in which the counts are expanded first to a 24-
highway agencies rationalize their traffic counting program, hour volume by using an hourly expansion factor, second to
especially when they have a tight budget. average daily volume by using a daily expansion factor, and
then to the annual flow by using a monthly expansion factor.
Keywords—Short period traffic counts, AADT estimation,
Obviously, the hourly factor would not be required when
Traffic monitoring, Traffic data, Factoring methodology,
coefficient of variation (COV))
counts are 24-hour or longer in duration. A more detailed
description of factor approach can be found in Garber and Hoel
(1999) [3].
I. INTRODUCTION
The past studies (Erhunmwunsee 1991[4], Phillips 1980[5],
Since the earliest days of commerce and transportation, Sharma 1983[6]) focused on the rationalization of counts that
humankind has been concerned with the numbers of people and were shorter than 24 hours in duration. They used data samples
things transported and the time and distance involved in from ATRs and evaluated them from the perspective of AADT
moving them. The subject of traffic data collection and analysis estimation. They used coefficient of variation (COV) of hourly
has been a concern from the earliest days of highway. Traffic expansion factor as a statistical measure of the expected
parameters estimated from such data are the foundation of reliability of the AADT estimate. Following are some of the
decision-makings related to the planning, design, operation and main conclusions of those studies:
management of highway transportation system. Due to the
critical roles of Annual Average Daily Traffic (AADT), it is • Longer counts are generally subject to smaller
important for highway agencies to get accurate AADT of each fluctuations in traffic volume, thus they are likely to
section of highways in their jurisdictions. The most reliable produce better estimates of AADT. However, in some
approach to obtain the AADT values is to install automatic cases, carefully selected schedules of shorter counts can
traffic recorders (ATRs) at all sections of each road in the results in lower COV than longer counts
networks. However, this is not a practical approach because of (Erhunmwunsee 1991).
the expense involved in purchasing, installing and maintaining
ATRs. So estimating AADT from sample traffic counts has • In general, counts with a duration of less than 12 hour
been universally accepted. yield lowest COV when their midpoint is centered at a
time between 2:00 pm and 4:00 pm. The improvement

250
978-1-4799-8694-1/11/31.00 © 2015 IEEE
The 3rd International Conference on Transportation Information and Safety, June 25 – June 28, 2015, Wuhan, P. R. China

from extending a count much beyond 6 hours is small Where

(Phillips 1980).
vj = sample daily volume of day j
• The sample count variation is a function of the nature of
n = total numbers of counts in the sample at the site
road site surveyed. The minimum variation in COV can
be expected for commuter sites and the maximum in the average value of sample volume vj
case of highly recreational routes (Sharma 1983).
SD = standard deviation of the volume variable vj
Drusch (1966) [7] evaluated the method advocated by U.S.
Bureau of Public Roads for estimating AADT from short-term COV = the coefficient of variation of sample volume (vj),
traffic counts and whether existing procedures could be %
improved with reduced annual cost. Tests were made using 7- The COV, as defined in Equation 2, is the standard
day, 5-day, 48-hour and 24-hour coverage counts. The study deviation expressed as a fraction (or percentage) of the mean
concluded that no significant difference between the accuracy value ¯v. A low value of COV, which is associated with less
of 7-day and 5-day coverage counts could be expected and a dispersion of individual sample volume (vj) about their mean,
significant increase in accuracy of estimates of AADT would reflects a high accuracy in estimating the actual traffic
be gained by using 7-day or 5-day coverage counts rather than volumes. In the study, the COV of different counting durations
48-hour or 24-hour coverage counts. for total traffic at study sites were investigated. Abbreviations
and Acronyms
III. STUDY DATA AND APPROACH
A total of 2 permanent traffic count (ATRs) sites location IV. RESULTS AND DISCUSSION
on Highway 16 in the Province of British Columbia, Canada, Initial investigations indicated that COV patterns for
were investigated. Traffic data for the year of 2013 were used various durations (1-day, 2-day, 3-day and 5-day) were
for the purpose of this study. generally similar at the investigation sites (Figure 1 and 2).
In the factoring method used for estimating AADT, all the Examination of Figure 1 and 2 reveals several important
expansion factors are averages. The true factors for any facts that are common to each site. In general, longer counts
individual count are almost certain to be different from the result in lower COV values. The lower COV values indicate
average factors. The more individual factors differ from smaller day to day fluctuations in the volume. Thus longer
average, i.e. the more variable they are, the greater is the likely counts are likely to produce better estimates of AADT. There is
error in the estimated annual flow. The most important cause of no indication that certain months always have lower COV
variability is site-to-site differences. This effect can be reduced values. Therefore, the month when the short counts were
by classification of the road sites. One of the other causes of conducted may not have a significant impact on the accuracy of
variability is the type of day. The fluctuation between weekday AADT estimation, but February, April, August and September
and weekend traffic counts tends to be considerable, therefore, tend to have lower COVs than the other months.
the weekend counts were not used in the accuracy analysis of
short period traffic counts; in other words, all samples for The effect of days on COV was also tested in the study.
accuracy analysis of short period traffic counts in this study Figure 3 and 4 indicated that there is no significantly difference
were generated from weekday data. The statutory holiday in the accuracy of AADT estimates from Monday to Friday,
counts were also excluded from the analysis. however the COV values for Monday, Tuesday and
Wednesday are 1-5% lower than the values for Thursday and
One of the several sources of error in the traditional method Friday.
of AADT estimation is the random daily variation in the traffic 25.00%

volume during any specified schedule of a short-period count. 20.00%

15.00%

The count with the lowest daily variation during the sampling
10.00%

5.00%

0.00%
Jan Feb Mar Apr May Jun J ul Aug Sep Oct Nov Dec

period can be expected to result in the lowest error of volume

estimation. In the tests of this study, a relative measure of
deviation, coefficient of variation (COV), was used in order to
compare the variation in several sets of data for short traffic
counts with different durations. COV was defined as follows:

( v −v )
2

¦
n
j =1 j
SD = (1) α + β = χ. (1) (1)
n −1

SD Fig. 1. COV Varies from Month by Month at Port Edward Road Site
COV = 100 × (2) α + β = χ. (1) (1)
v

251
978-1-4799-8694-1/11/31.00 © 2015 IEEE
The 3rd International Conference on Transportation Information and Safety, June 25 – June 28, 2015, Wuhan, P. R. China

V. CONCLUSIONS AND RECOMMENDATIONS

14.00%

12.00%

10.00%

8.00%

6.00%

The results of this study reveal that the accuracies of AADT

4.00%

2.00%

0.00%
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

estimates for short period traffic counts vary from day to day
and month to month. The best of months for conducting short
counts in order to produce accurate AADT estimates are site
specific. Local knowledge may play an important role to
determine the time of conducting short period traffic counts.
The short period traffic counts surveyed in the early of the
week may provide better AADT estimates than in the late of
the week. The traffic counts surveyed in the early of the week
(i.e., Monday to Wednesday) may produce more accurate
AADT estimates than the counts from late of the week
(Thursday and Friday).
In general, traffic counts with longer duration will produce
better AADT estimates. In the study sites, the COV values for
Fig. 2. COV Varies from Month by Month at Bednesti Site 5-day counts were 5-9% when the COV values for 1-day
counts were 6-12%. However, the difference is not significant
14 .00%

13 .00%

12 .00%

11.00%
between the accuracy of 5-day and 1-day traffic counts. The
accuracy of 1-day and 2-day traffic counts is very similar. A
10 .00%

9 .00%

8 .00%

7 .00%

practical implication of such a finding is that shorter counts (1

6 .00%
Mon Tue Wed Thu Fri

or 2-day counts) conducted between Monday and Wednesday

may produce AADT estimates as good as the estimates from
longer counts (5-day counts). The analysis and findings of this
study on the adequacy of short period traffic counts from the
perspective of AADT estimation will help highway agencies
rationalize their traffic counting program, especially when they
have a tight budget. The study results clearly suggest that some
existing procedures (5-day counts) could be changed for a
reduced annual cost.

REFERENCES

Fig. 3. Effect of Days on COVs at Port Edward Road Site [1] Traffic Monitoring Guide (draft), U.S. Department of Transportation,
FHWA, Washington, DC, 2001.
[2] Sharma, S. C., P. Lingras, F. Xu and G. Liu, Neural Networks as
Alternative to Traditional Factor Approach of Annual Average Daily
Traffic Estimation from Traffic Counts. Transportation Research Record
10.00%
1660, TRB, National Research Council, Washington, DC, 1999.
COV of Sample Volume (vj)

9.00%
[3] Garber, N.J. and Hoel, L.A., Traffic and Highway Engineering.
8.00% 1-day Brooks/Cole Publishing Company, Pacific grove, CA, USA, 1999.
7.00% 3-day [4] Erhunmwunsee, O. Paul, Estimating Average Daily Traffic Flow from
Short Period Counts. ITE Journal, pp. 23-30, November 1991.
6.00% 2-day
5-day [5] Philips, Garwyn, When to Mount a traffic Count. Traffic Engineering
5.00% and Control, January 1980.
4.00% [6] Sharma, Satish C., Minimizing Cost of Manual Traffic Counts:
3.00% Canadian Example. Transportation Research Record 905, TRB, National
Research Council, Washington, DC, 1983.
2.00%
[7] Drusch, Robert L., Estimating Annual Average Daily Traffic Flow from
1.00% Short-term Traffic Counts. Highway Research Record, Vol. 118, 1966.
0.00%
Mon Tue Wed Thu Fri
Mid-point of Count (day)

Fig. 4. Effect of Days on COVs at Bednesti Site

252
978-1-4799-8694-1/11/31.00 © 2015 IEEE