estimating annual average daily traffic (aadt) for local ...docs.trb.org/prp/13-3490.pdf · 3 for...
TRANSCRIPT
1
Estimating Annual Average Daily Traffic (AADT) 2
for Local Roads for Highway Safety Analysis 3
4
5
by 6
7
8
Tao Wang, Ph.D. 9 Research Associate 10
Lehman Center for Transportation Research 11
Florida International University 12
10555 West Flagler Street, EC 3680 13
Miami, FL 33174 14
Phone: (305) 510-3331 15
Email: [email protected] 16
17
Albert Gan, Ph.D.* 18 * Corresponding Author
19
Professor 20
Department of Civil and Environmental Engineering 21
Florida International University 22
10555 West Flagler Street, EC 3680 23
Miami, FL 33174 24
Email: [email protected] 25
26
Priyanka Alluri, Ph.D. 27 Research Associate 28
Lehman Center for Transportation Research 29
Florida International University 30
10555 West Flagler Street, EC 3680 31
Miami, FL 33174 32
Email: [email protected] 33
34
35
This paper contains 5,000 words, 7 figures, and 2 tables. 36
(7,250 equivalent words) 37
38
39
Submitted for: 40
Presentation and Publication 41
The 92nd Annual Meeting of the Transportation Research Board 42
43
44
November 15, 2012 45
46
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 1
ABSTRACT 1 2
Annual average daily traffic (AADT) is a required input to the newly released SafetyAnalyst 3
software application. Further, AADT is also required to calculate crash rates. Traditionally, 4
AADTs are estimated using a mix of permanent and temporary traffic counts collected in the 5
field. Because field collection of traffic counts is expensive, it is usually performed for only the 6
major roads. The mandate by the Federal Highway Administration (FHWA) to report the top 5% 7
of high crash locations on all public roads underscores, for the first time, the need for state 8
Departments of Transportation to acquire AADTs for also the non-state local roads. However, 9
many local jurisdictions either do not have any AADT data, or they do not have them in 10
sufficient quality. This paper presents a method to estimate AADTs for local roads using the 11
travel demand modeling method. A major component of the method involves a parcel-level trip 12
generation model that estimates the trips generated by each parcel. The generated trips are then 13
distributed to existing traffic count sites using a parcel-level trip distribution gravity model. The 14
all-or-nothing trip assignment method is then applied to assign the trips between the parcels and 15
the traffic count sites onto local roadway network to yield estimates of AADTs. The estimated 16
AADTs were compared with those from an existing regression-based method using actual traffic 17
counts from Broward County, Florida. The results show that the proposed method produces 18
significantly lower mean absolute percentage errors. 19
20
Key words: Annual Average Daily Traffic, Tax Parcel Records, Trip Generation, Travel 21
Demand Modeling. 22
23 24
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 2
INTRODUCTION 1 2
Annual average daily traffic (AADT) is the average 24-hour traffic volume at a roadway location 3
over an entire year. AADT is required for many transportation analyses including economic 4
evaluation of highway safety projects, estimation of highway user revenues, computation of 5
highway statistics such as vehicle miles traveled (VMT), development of improvement and 6
maintenance programs, etc. 7
One well-known application of AADT is in the calculation of crash rates. As part of the 8
new Highway Safety Improvement Program (HSIP) of the Federal Highway Administration 9
(FHWA), states are required to submit an annual report describing no less than 5% of their 10
highway locations on all public roads that exhibit the most severe safety needs (1). To submit the 11
annual 5% reports (also known as Transparency Report), the Florida Department of 12
Transportation (FDOT) needs to have AADTs for all roads in Florida; however, FDOT estimated 13
AADTs for only its state roads but not the local roads. 14
In addition to the need for 5% reporting, AADT is also a required input to SafetyAnalyst 15
(2), a new safety analysis software released by the American Association of State Highway and 16
Transportation Officials (AASHTO). The software aims at providing state and local highway 17
agencies with a comprehensive set of tools to enhance their programming of site-specific 18
highway safety improvements. FDOT is interested in deploying SafetyAnalyst for all roads in 19
Florida. Because AADT is a required input to SafetyAnalyst, FDOT was, again, faced with the 20
problem of not having AADTs for local roads. 21
The most accurate method for obtaining the AADT of a roadway segment is to install an 22
Automatic Traffic Recorder (ATR) to count the total volumes continuously throughout the entire 23
year. However, because the installation and maintenance of permanent counters are expensive, 24
the number of permanent counters is typically limited. Therefore, it is not economically feasible 25
to apply this method of AADT estimation on a widespread basis. 26
An alternative approach to estimating AADT is to use portable counts, also called short-27
term, seasonal, or coverage counts. The collected short-term volumes on the interested roads are 28
then used to calculate Average Daily Traffic (ADT) which is then converted to AADT by 29
applying some adjustment factors. This factor approach is more economically feasible than the 30
permanent count method, but is still too costly to cover the local roads, which number over two 31
million segments in the state of Florida. 32
In 2007, FDOT contracted with the University of South Florida to develop regression 33
models to estimate AADTs for the local roads (3). The project attempted to improve upon a set 34
of regression models developed earlier by Xia et al. (4). However, a preliminary evaluation 35
based on data from the Miami-Dade and Broward counties, Florida showed that the estimation 36
errors of the models exceeded 100% and 200%, respectively. 37
This paper presents an improved method of estimating AADTs for local roads in Florida. 38
The method applies travel demand modeling techniques at the tax parcel level. In this method, 39
trips are generated based on parcel data using the trip generation rates from the trip generation 40
report of the Institute of Transportation Engineers (ITE) (5). A gravity trip distribution model is 41
then developed to distribute the parcel trips to the traffic counts sites on the major roads. All the 42
trips are finally assigned to local roads by applying all-or-nothing assignments based on free-43
flow travel times. 44
45
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 3
LITERATURE REVIEW 1 2
Several existing AADT estimation methods have been reported in the literature. Among the 3
methods the regression analysis has been the most widely used. A non-exhaustive list of states 4
that has developed regression models to estimate AADTs for their highway networks would 5
include Kentucky (6), Alabama (7), Minnesota (8), and Indiana (9). In Florida, regression models 6
were first developed to estimate AADTs for local roads in Broward County (4). The models were 7
modified and improved in multiple follow-up studies (10, 11, 12). A more recent attempt to 8
estimate AADTs for local roads in Florida was performed by Lu et al. (3). As aforementioned, 9
they attempted to improve the regression models developed by Xia et al. (4). However, the 10
estimation errors from their models were found to be quite high. 11
The availability of high-resolution satellite images and aerial photos had encouraged 12
some researchers to explore the potential of estimating AADTs with image-based data. McCord 13
et al. (13) first proposed an image-based method for AADT estimation using satellite images and 14
aerial photos. Later, a related method combining image-based and ground-based estimations was 15
implemented for Ohio road segments (14, 15). One problem with image-based methods is that it 16
is difficult to estimate traffic volumes on local roads as traffic on such facilities is usually sparse 17
and infrequent. 18
Another method attempted for AADT estimation was based on machine learning 19
algorithms. For example, a multi-layered, feed-forward, and back-propagation neural network 20
with supervised learning was developed to compare the Artificial Neural Network (ANN) 21
approach for AADT estimation to the traditional factor approach with 48-hour short-term count 22
data (16). The ANN approach was later applied by the same researchers to estimate AADTs on 23
low-volume rural roads (17, 18). Another effort that involved ANN models for AADT estimation 24
was conducted by Lam and Xu (19). A more recent study was performed by Li et al. (20) who 25
combined a K-Nearest Neighbor (K-NN) algorithm with Geographic Information System (GIS) 26
technology to aid in AADT estimation. In another recent study, a modified version of support 27
vector regression machine (SVR) named SVR-DP (SVR with Data-dependent Parameters) was 28
used to forecast AADT one year into the future based on historical AADTs (21). 29
Studies using the travel demand modeling approach for AADT estimation are relatively 30
rare. One known study that used this approach was conducted by Zhong and Hanson (22). The 31
study developed travel demand models to estimate AADTs on low-class roads for two regions in 32
the province of New Brunswick in Canada (22). While the study showed that the method has the 33
potential to improve AADT estimation, the results from the study were negatively impacted by 34
the available census geographic unit data, called dissemination area (DA), which was the 35
smallest census unit available in Canada. However, a DA by definition is a geographic unit 36
composed of one or more neighboring dissemination blocks with a population of 400 to 700 37
persons. Accordingly, data based on DAs are clearly too coarse for applications involving local 38
roads, which typically attract most trips generated from the abutting land uses. In this research, 39
this travel demand modeling approach is applied based on data at the parcel level. 40
41
METHODOLOGY 42 43
A parcel-level travel demand analysis model was implemented in this study. Similar to the 44
traditional travel demand forecasting model, the parcel-level model also involves a series of 45
mathematical models to simulate human travel behavior. The model consists of the following 46
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 4
four steps: network modeling, parcel-level trip generation, parcel-level trip distribution, and 1
parcel-level trip assignment. 2
Different from the traditional travel demand model, which attempts to simulate the 3
choices that the travelers may make during the entire trip from the origination to the destination, 4
the parcel-level model attempts to simulate choices that travelers may make in response to the 5
given local street system. The model does not include the mode choice step as the transit trips 6
and other modes are insignificant on local roads. The functionalities of each step involved in the 7
parcel-level travel demand analysis model are as follows: 8
Network Modeling defines the boundaries of the study area, prepares and preprocesses 9
the roadway network, parcel, and traffic counts data, and sets up the network 10
representation of the roadway linked with parcels and traffic count sites. 11
Parcel-level Trip Generation estimates the number of vehicle trips generated by each 12
parcel in the study area. The estimation is calculated based on the land use type of each 13
parcel and its corresponding ITE trip generation rate. 14
Parcel-level Trip Distribution determines where the trips generated by each parcel will 15
go. It determines the number of trips between a parcel and a traffic count site based on 16
traffic count data (or AADT estimated from the count data) and the shortest travel time 17
between them. 18
Parcel-level Trip Assignment predicts the routes the travelers will take to reach the traffic 19
count sites on major roads, resulting in the estimated AADTs of local roads in the study 20
area. 21
In the network modeling step, the study area boundary, commonly called the cordon line, 22
is defined. Unabridged roadway network data, detailed parcel data, and traffic count sites data 23
were used in the analysis. Similar to the traditional travel demand model, centroids and centroid 24
connectors were created. Each parcel was assigned a centroid and a centroid connector to 25
represent access to a neighboring road segment. An example of a subarea with the added centroid 26
connectors connecting the parcels and the closest roads is shown in Figure 1. The green polygons 27
represent the parcel boundaries; the blue lines represent the roadway; and the gray lines are the 28
added centroid connectors. 29
Parcel-level trip generation is the process to estimate and quantify the number of trips 30
each parcel will generate. In this application, this step was implemented using both the 31
Department of Revenue (DOR) parcel data and the trip generation rates and regression equations 32
provided by ITE Trip Generation Report (5). 33
The DOR parcel data has 100 land use types for the parcels, and the ITE Trip Generation 34
Report provides trip generation rates and/or equations for 10 main land use categories and 162 35
sub-categories. The two types of land use categories were matched to estimate the parcel trips. In 36
the case that a parcel land use type encompasses multiple ITE land use categories, the results 37
were averaged and weighted by an estimate of the relative presence of each ITE land use 38
category in the study area. The calculation can be expressed as follows: 39
40
ii Pn
i
XFT
)
1
( (1) 41
where, 42
T = number of vehicle trips generated by a parcel; 43
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 5
Fi = ITE trip generation function (either regression equation or average trip rate) for 1
ITE land use type i; 2
X = independent variable such as dwelling units, gross floor area, etc.; 3
n = number of ITE land use types; and 4
Pi = proportion of ITE land use type i in the study area. 5
6
7 8
FIGURE 1 Example of study area with parcels (green lines), centroid connectors (gray 9
lines), and existing roads (blue lines). 10 11
For most of the land use types, the ITE trip generation rates/equations are based 12
on dwelling units or areas, which are also the attributes of a parcel in the parcel database. Hence, 13
for a majority of parcels, the trip generation can be calculated directly by using the parcel data. 14
However, for land use types of which the corresponding independent variable used by ITE rates 15
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 6
is a size attribute that is different from the parcel data, they are adjusted by the ratio between the 1
ITE and parcel based mean values using the following equation: 2
3
)( RXFT (2) 4
5
where, 6
T = number of vehicle trips generated by a parcel; 7
F = the ITE trip generation function (either regression equation or average trip rate, 8
which can be regarded as a special linear regression); 9
X = independent variable such as dwelling units, gross floor area, etc.; and 10
R = the ratio of mean values between the ITE independent variable and the parcel 11
size attribute. 12
13
Due to the lack of other demographic and land use data for a few uncommon land use 14
types, the proportion of ITE land use type Pi in Equation (1) and the ratio of mean values 15
between the ITE independent variable and the parcel size attribute R in Equation (2) were 16
determined by human judgment. 17
Due to the different travel patterns among the weekday, Saturday, and Sunday, the 18
number of trips for weekdays and weekend were averaged using the following equation: 19
20
7
5 SundaySaturdayweekday
average
TTTT
(3) 21
where, 22
Taverage = the final estimated number of daily trips generated by a parcel, 23
Tweekday = average weekday trips generated by a parcel, 24
TSaturday = Saturday trips generated by a parcel, and 25
TSunday = Sunday trips generated by a parcel. 26
27
Similar to the traditional travel demand model, parcel-level trip distribution is also 28
derived from Newton's law of gravity. However, the trips are distributed between the parcels and 29
the traffic count sites instead of among the zones. To distribute trips, all the parcels have only 30
productions and no attractions, and all the traffic count sites have only attractions and no 31
productions. Productions of a parcel are the trips generated in the parcel-level trip generation 32
step, and the attractions of a traffic count site are either the traffic count data or AADT estimated 33
from the traffic count data. In addition, while traditional travel demand model usually includes 34
the effects of multiple travel impedance factors, such as travel time, cost, etc., parcel-level trip 35
distribution considers only the shortest free-flow travel time. This is expedient because travel 36
time is the major factor that determines trips on local roads, and travelers will most likely choose 37
the fastest path to access major roads to reach their destinations. The parcel-level trip distribution 38
can be expressed as follows: 39
40
i
ikk
ijj
ij Tn
k
DA
DAT
1
)/(
/ (4) 41
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 7
where, 1
Tij = daily vehicle trips between parcel i and traffic count site j, 2
Ti = total daily vehicle trips generated by parcel i, 3
Aj = AADT estimated from traffic count volume at traffic count site j, 4
Dij = the shortest free flow travel time between parcel i and traffic count site j, and 5
n = number of traffic count sites. 6
7
It should be noted that, since the parcel-level trip distribution step distributes the trips 8
from the parcels to the nearby traffic count sites, enough traffic count data are required to evenly 9
cover the entire study area. Uneven coverage of traffic count sites may cause inaccurate 10
distribution of trips. 11
For traffic assignment, the traditional travel demand model commonly assigns trips using 12
an all-or-nothing with capacity restraint method, also known as the equilibrium assignment 13
method. In this application, only the simple all-or-nothing assignment method is needed because 14
congestion seldom happens on local roads. In other words, for the same reason that affects the 15
trip distribution step, the travelers’ route selection is based mainly on the free-flow travel time to 16
the nearby major roads. The simple all-or-nothing assignment also improved the efficiency of the 17
model as it involves only one iteration. 18
After the trips for all of the parcels were assigned to the local roads, the AADT is 19
estimated as the sum of trips in both the directions for a roadway segment. Finally, data post-20
processing was performed to calculate the final AADT values for all local roads. 21
22
MODEL DEVELOPMENT 23 24
Two development tools were applied to implement this model: ArcGIS 10.0 from ESRI and 25
Cube 5.1.3 from Citilabs. An application of ArcGIS called ModelBuilder was used to perform 26
the data preprocessing and post-processing. Scripts were then developed in Cube Voyager to call 27
and customize the library programs to implement the model. Figure 2 shows the system 28
components and procedure used to estimate AADT. The procedure includes the following sub-29
steps: 30
ArcGIS is used to preprocess the input data for the model including the DOR parcel data, 31
unabridged highway network data, and traffic count sites data. 32
The preprocessed input data are imported into Cube, and the highway network is built 33
from the unabridged roadway shape file. 34
The built highway network is used by the network modeling step to calculate the free 35
flow travel time skim matrix. 36
The parcel-level trip generation step is performed by using the merged DOR parcel data 37
and traffic count site data as well as the trip generation rates and regression equations 38
provided by the ITE Trip Generation Report. 39
A parcel-level trip distribution gravity model is used to distribute the generated trips 40
between the parcels and the nearby traffic count sites. 41
The distributed trips are assigned to the network using all-or-nothing assignment method 42
in the parcel-level trip assignment step. 43
The traffic volume data of the loaded network are exported, and ArcGIS is used again to 44
calculate the final AADTs, which are then joined to the original roadway network to 45
obtain the roadway network with AADTs. 46
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 8
1
2 FIGURE 2 System components and procedure. 3
4
ArcGIS ModelBuilder, which provides a visual programming environment allowing users 5
to graphically link geoprocessing tools into models, was used to implement the ArcGIS 6
component. While the models built with ModelBuilder can be executed directly in ArcGIS, they 7
can also be exported to scripting language such as Python. The Python scripts can be called with 8
the Cube Voyager Pilot program, so theoretically all the steps of the ArcGIS part can be 9
integrated into Cube to simplify the execution of the entire model. However, because this part of 10
the procedure requires some geo-processing tools that are supported only by ArcGIS 10.0, which 11
is not compatible with the current version of Cube (5.1.3), integration of ArcGIS into Cube could 12
not be realized. Nevertheless, this incompatibility would not affect the results of the entire 13
model. 14
Figure 3 shows the model steps and the input and output files for each steps implemented 15
in Cube. When the model is run, the four steps are executed in sequence, and the output files of a 16
preceding step become the input files of a later step. It is noted that if there were no compatibility 17
problems as mentioned above, the ArcGIS part of the procedure could have been combined with 18
Cube, and the steps shown in Figure 3 would be all the steps involved in the entire model. 19
Table 1 summarizes the input and output files for each step. There are two input files for 20
Cube: the network file preprocessed by ArcGIS, and the DBF file for the merged parcels and 21
traffic count sites shape file which is also generated by ArcGIS. There is one output file 22
generated by the Cube part, i.e., the DBF link file with the traffic volume information exported 23
from the loaded network assigned in the parcel-level trip assignment step. Among the steps, the 24
output files of a preceding step become the input files of a later step. 25
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 9
1 FIGURE 3 Model steps in Cube. 2
3
TABLE 1 Input and Output Files 4
Model Step Input File Output File
Network Modeling Preprocessed Network File Free Flow Time Skim Matrix File
Modified Network File
Parcel-level Trip
Generation Merged Parcels and Counts DBF File Vehicle Trips DBF File
Parcel-level Trip
Distribution Free Flow Time Skim Matrix File
Distributed Trips Matrix File Vehicle Trips DBF File
Parcel-level Trip
Assignment Distributed Trips Matrix File Link DBF File with Volume Exported
from Loaded Network Modified Network File
5
MODEL EVALUATION 6 7
Because the actual AADT values for local roads were not available, the AADTs estimated from 8
short-term traffic count data had to be used as the ground truths for model evaluation in this 9
study. To quantify the estimation error, the mean absolute percentage error (MAPE), calculated 10
as follows, was used: 11 12
n
i iG
iGiF
nMAPE
1 )(
)()(1
(5)
13
14 where, 15
G(i) = ground truth AADT at location i, 16
F(i) = estimated AADT at location i, and 17
n = total number of locations. 18
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 10
As aforementioned, regression modeling has been the most widely used AADT 1
estimation method. Based on the method’s popularity and also the availability of the 2
experimental results, the regression method proposed by Lu et al. (3) was compared with the 3
proposed parcel-level travel demand model. Lu et al. developed multiple regression models that 4
estimate AADTs as functions of socio-economic and roadway characteristics. 5
Broward County in Florida was selected as the study area for evaluation. The County 6
includes over 4,000,000 parcels. However, Cube 5.1.3 can only process a maximum of 32,000 7
zones (or parcels in this application) at a time. A total of 10 subareas with a total of 78 evaluating 8
count sites were selected to cover diverse areas in this evaluation. An example map of the 9
subareas is shown in Figure 4. Two types of count sites are shown in this map. One is the 10
estimation count sites on the major roads (indicated by circular dots in the map) which are used 11
to estimate the AADTs, and the other is evaluation count sites on the minor roads (indicated by 12
the triangles in the map) which are used to evaluate the estimation results. 13
14
15
FIGURE 4 An example of the subareas. 16
Figures 5 and 6 compare the overall performance of the regression method and the 17
proposed method, respectively. As shown in the figures, the maximum AADT is lower than 18
30,000 vehicles/day and within a reasonable range, since all the testing locations were on local 19
roads. From Figure 5, it can be seen that the regression models tend to overestimate AADT. The 20
plotted dots in Figure 6 show that the results of the proposed method are distributed more evenly 21
on both sides of the 45-degree sloping line than in Figure 5. 22
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 11
1 2
FIGURE 5 Comparison of estimated AADTs from regression models and ground truth 3
AADTs. 4
5
6
7 8
FIGURE 6 Comparison of estimated AADTs from parcel-level demand model with ground 9
truth AADTs. 10 11
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 12
Table 2 quantifies the accuracy of the two methods using the MAPE measure. It can be 1
seen that the proposed method has consistently lower estimation error than the regression models 2
for all study areas. The overall estimation errors for all the 78 evaluating count sites in the 10 3
study areas were also calculated. The results show that the proposed model generated 52% 4
MAPE, which is 159% lower than the corresponding value of 211% from the regression models. 5
6
TABLE 2 Errors Comparison of Regression and Demand Models 7
Subarea # MAPE from Regression Models (%) MAPE from Demand Model (%)
1 217 48 2 314 52 3 177 48 4 345 66 5 1756 49 6 405 60 7 114 62 8 186 49 9 181 56
10 157 39 All Subareas 211 52
8
Depending on the availability of traffic count data, most of the traffic count sites used for 9
this evaluation are located on local roads that are directly connected to the state roads. The 10
lower-level local roads such as the community roads were not used in this evaluation because of 11
the lack of traffic count data. However, the proposed method is expected to perform better even 12
for lower-level local roads as its trip generation is based on detailed parcel level data. To verify 13
this assumption, the AADTs estimated using the two methods for the available lower-level local 14
roads were checked and compared. Figure 7 shows the estimation results for the roads in a 15
community of approximately 160 houses. In this figure, the AADTs estimated from the proposed 16
method and the regression method are displayed in red and green, respectively. It can be seen 17
that the AADTs estimated by the regression method were unreasonably high. One reason that the 18
regression method significantly overestimated the AADTs in this case was due to their inability 19
to recognize that the layout of the local roads was designed such that through traffic, if any, was 20
minimal. 21
22
CONCLUSIONS 23 24
Annual average daily traffic (AADT) data are needed to identify high crash locations on all 25
public roads for 5% reporting as mandated by FHWA. They are also required for input to the 26
newly released SafetyAnalyst software application. The lack of AADT data, especially for the 27
vast local road networks, prevents such application from being deployed to improve safety on 28
local roads. This paper described a parcel-level travel demand model that was developed to 29
estimate AADTs for local roads. Compared to results from a set of existing regression models, 30
the parcel-level demand model improved the accuracy of AADT estimation for local roads by 31
159% based on the MAPE measure. A major component of the proposed model involves a 32
parcel-level trip generation model that estimates the trips generated by each parcel. The model 33
makes use of the tax parcel data together with trip generation rates from the Institute of 34
Transportation Engineers (ITE) Trip Generation Report. One advantage of using tax parcel data 35
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 13
is that the data is updated at least annually, making it possible to update the AADTs in response 1
to lane use changes. One disadvantage of the proposed model is the large number of parcels that 2
have to be preprocessed to improve the model’s efficiency. For applications involving large 3
areas, the study area has to be divided into multiple subareas and run the model separately. In 4
addition, the model also requires that there not only be sufficient traffic count data to cover the 5
entire study area, but they should be relatively evenly spaced, as uneven coverage of traffic count 6
sites may result in inaccurate distribution of trips. 7
8
9 10
FIGURE 7 Example of AADT estimation for local roads in a community. 11 (Note: AADTs estimated by the proposed method and the regression method are shown in red and green, respectively.) 12
13
ACKNOWLEDGEMENTS 14 15
This research was funded by the Research Center of the Florida Department of Transportation 16
(FDOT). The authors are thankful to the Project Manager, Mr. Joseph Santos, of the FDOT 17
Safety Office for his guidance and support. 18
19
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 14
REFERENCES 1
2 1. Federal Highway Administration. "5 Percent Report" Requirement. 3
http://safety.fhwa.dot.gov/hsip/fivepercent/, Accessed April 15, 2012. 4
2. American Association of State Highway and Transportation Officials. SafetyAnalyst. 5
http://www.safetyanalyst.org/, Accessed April 20, 2012. 6
3. Lu, J., T. Pan, and P. Liu. Assignment of Estimated Average Annual Daily Traffic Volumes on 7
All Roads in Florida. Final Report Prepared for Florida Department of Transportation, 8
Tallahassee, FL, 2007. 9
4. Xia, Q., F, Zhao, L.D Shen, and D. Ospina. Estimation of Annual Average Daily Traffic for 10
Non-State Roads in a Florida County. Research Report Prepared for Florida Department of 11
Transportation, Tallahassee, FL, 1999. 12
5. Trip Generation: An ITE Informational Report, 8th Edition. Institute of Transportation 13
Engineers, Washington, D.C., 2008. 14
6. Deacon, J. A., J. G. Pigman, and A. Mohenzadeh. Traffic Volume Estimates and Growth 15
Trends. UKTRP-8732, Kentucky Transportation Research Program, University of Kentucky, 16
Kentucky, 1987. 17
7. Shon, D. Traffic Volume Forecasting for Rural Alabama State Highways. Thesis, Auburn 18
University, Alabama, 1989. 19
8. Cheng, C. Optimum Sampling for Traffic Volume Estimation. Ph.D. Dissertation, University 20
of Minnesota, Minnesota, 1992. 21
9. Mohamad, D., K.C. Sinha, T. Kuczek, and C.F. Scholer. An Annual Average Daily Traffic 22
Prediction Model for County Roads. In Transportation Research Record No. 1617, Journal 23
of the Transportation Research Board, Transportation Research Board, Washington, D.C., 24
1998, pp. 69-77. 25
10. Shen, L. D., F. Zhao, and D. Ospina. Estimation of Annual Average daily Traffic for Off-26
System Roads in Florida. Research Report Prepared for Florida Department of 27
Transportation, Tallahassee, FL, 1999. 28
11. Zhao, F. and S. Chung. Estimation of Annual Average Daily Traffic in a Florida County 29
Using GIS and Regression, In Transportation Research Record No. 1660, Journal of the 30
Transportation Research Board, Transportation Research Board, Washington, D.C., 1999, 31
pp. 32-40. 32
12. Park, N. Estimation of Average Annual Daily Traffic (AADT) Using Geographically 33
Weighted Regression (GWR) Method and Geographic Information System (GIS). Ph.D. 34
Dissertation, Florida International University, Miami, FL, 2004. 35
13. McCord, M. R., Y. Yang, Z. Jiang, B. Coifman, and P. Goel. Estimating AADT from 36
Satellite Imagery and Air Photographs: Empirical Results. In Transportation Research 37
Record No. 1855, Journal of the Transportation Research Board, Transportation Research 38
Board, Washington, D.C., 2003, pp. 136-142. 39
14. Jiang, Z., M. R. McCord, and P. K. Goel. Improved AADT Estimation by Combining 40
Information in Image- and Ground-Based Traffic Data. Journal of Transportation 41
Engineering, Vol 132, No. 7, 2006, pp. 523-530. 42
15. Jiang, Z., M. R. McCord, and P. K. Goel. Empirical Validation of Improved AADT 43
Estimates from Image-Based Information on Coverage Count Segments. In TRB 86th Annual 44
Meeting Compendium of Papers. CD-ROM. Transportation Research Board, Washington, 45
D.C., 2007, pp. 15. 46
TRB 2013 Annual Meeting Paper revised from original submittal.
Wang, Gan, and Alluri 15
16. Sharma, S. C., P. Lingras, F. Xu, and G. X. Liu. Neural Networks as Alternative to 1
Traditional Factor Approach of Annual Average Daily Traffic Estimation from Traffic 2
Counts. In Transportation Research Record No. 1660, Journal of the Transportation 3
Research Board, Transportation Research Board, Washington, D.C., 1999, pp. 24-31. 4
17. Sharma, S. C., P. Lingras, G. X. Liu, and F. Xu. Application of Neural Networks to Estimate 5
AADT on Low Volume Roads. In Transportation Research Record: Journal of the 6
Transportation Research Board, No. 1719, Transportation Research Board, Washington, 7
D.C., 2000, pp. 103-111. 8
18. Sharma, S. C., P. Lingras, F. Xu, and P. Kilburn. Application of Neural Networks to Estimate 9
AADT on Low-volume Roads. ASCE Journal of Transportation Engineering, Vol. 127, No. 10
5, 2001, pp. 426-432. 11
19. Lam, W. H. K. and J. Xu. Estimation of AADT from Short Period Counts in Hong-Kong – A 12
Comparison between Neural Network Method and Regression Analysis. Journal of 13
Advanced Transportation, Vol 34, No. 2, 2000, pp. 249-268. 14
20. Li, J., C. Xu, and J. D. Fricker. Comparison of Annual Average Daily Traffic Estimates: 15
Traditional Factor, Statistical, Artificial Neural Network, and Fuzzy Basis Neural Network 16
Approach. In TRB 87th Annual Meeting Compendium of Papers. CD-ROM. Transportation 17
Research Board, Washington, D.C., 2008, pp. 19. 18
21. Castro-Neto, M., Y. Jeong, M. K. Jeong, and L. D. Han. AADT Prediction Using Support 19
Vector Regression with Data-dependent Parameters. Expert Systems with Applications, Vol. 20
36, No. 2, 2009, pp. 2979-2986. 21
22. Zhong, M. and B. L. Hanson. GIS based Travel Demand Modeling for Estimating Traffic on 22
Low-class Roads, Transportation Planning and Technology, Vol. 32, No. 5, 2009, pp. 423-23
439. 24
TRB 2013 Annual Meeting Paper revised from original submittal.