xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 002000378
007 cr mnu|||uuuuu
008 090421s2008 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002481
Assignment of estimated average annual daily traffic on all roads in Florida
h [electronic resource] /
by Tao Pan.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 87 pages.
Thesis (M.S.C.E.)--University of South Florida, 2008.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
ABSTRACT: In the first part, this thesis performed a study to compile and compare current procedures or methodologies for the estimation of traffic volumes on the roads where traffic counts are not easily available. In the second part, linear regression was practiced as an AADT estimation process, which was primarily based on known or accepted AADT values on the neighboring state and local roadways, population densities and other social/economic data. To develop AADT prediction models for estimating AADT values, two different types of database were created, including a social economic database and a roadway characteristics database. Ten years social economic data, from 1995 to 2005 were collected for each of the 67 counties in the state of Florida, and a social economic database was created by manually imputing data obtained from different resources into the social economic database. The roadway characteristics database was created by joining different GIS data layers to the Tele Atlas base map provided by Florida Department of Transportation (FDOT). Stepwise regression method was used to select variables that will be included into the final models. All selected independent variables in the models are statistically significant with a 90% level of confidence. In total, six linear regression models were built. The adjusted R2 values of the AADT prediction models vary from 0.166 to 0.418. Model validation results show that the MAPE values of the AADT prediction models vary from 31.99% to 159.49%. The model with the lowest MAPE value is found to be the minor state/county highway model for rural area. The model with the highest MAPE value is found to be the local street model for large metropolitan area. In general, minor state/county highway models provide more reasonable AADT estimates as compared to the local street model in terms of the lower MAPE values.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Advisor: Jian Lu
x Civil & Environmental Engineering
t USF Electronic Theses and Dissertations.
Assignment of Estimated Average Annual Daily Traffic Volumes on All Roads in Florida by Tao Pan A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Civil Engineering Department of Civil & Environmental Engineering College of Engineering University of South Florida Co-Major Professor: Jian Lu, Ph.D. Co-Major Professor: Pan Liu, Ph.D. Edward Mierzejewski, Ph.D. Pei-Sung Lin, Ph.D. Manjriker Gunaratne, Ph.D. Date of Approval: March 27, 2008 Keywords: AADT, Linear Regression, Soci al Economy, Traffic Count, Database Copyright 2008, Tao Pan
DEDICATION This thesis is dedicated to my parents who have support me all the way since the beginning of my studies.
ACKNOWLEDGMENTS It is with great pride that I thank the br illiant minds affiliated with the Department of Civil and Environment Engineering at the Un iversity of South Flor ida. I would like to give special thanks to my major professor, Dr. Jian John Lu, for the guidance he has provided. In addition, I would like to thank committee members Dr. Edward Mierzejewski and Dr. Pei-Sung Lin. This thes is would not have been possible without your contributions.
NOTE TO READER The original of this document contains color that is necessa ry for understanding the data. The original thesis is on file with the USF library in Tampa, Florida.
i Table of Contents List of Tables iii List of Figures v Abstract vii Chapter One 1 1.1Background Information 1 1.2 Research Objective and Approaches 5 1.3 Organization of the Thesis 5 Chapter Two 7 2.1 Introduction 7 2.2 Traffic Count 7 2.3 Traffic Count Program 8 2.4 AADT Conversion with ATR Data 9 2.5 AADT Conversion with C overage Count Data 11 2.6 AADT Estimation Models 14 2.7 Other AADT Estimation Methods 21 2.8 Summary 23 Chapter Three 24 3.1 Introduction 24 3.2 Methods for Assigning AADT Values to Type I Streets 28 3.3 Methods for Assigning AADT Values to Type II Streets 29 Chapter Four 32 4.1 Introduction 32 4.2 Roadway Characteristics Database 32 4.3 Social Economic Database 34 Chapter Five 36 5.1 Model Calibration 36 5.1.1 Introduction 36 5.1.2 Variable Description 36 5.2 Model Development 41 5.3 Model Validation 46 5.3.1 Model Validation for Large Metropolitan Area 48
ii 5.3.2 Model Validation for Small-Medium Urban Group 50 5.3.3 Model Validation for Rural Area Group 50 Chapter Six 56 6.1 Summary 56 6.2 Final Result 58 References 61 Bibliography 63 Appendices 6 4 Appendix A: Descriptive Stat istics for Traffic Counts 65 Appendix B: Type I Road Assignment 68 Appendix C: Type III Roads 73 Appendix D: County Group 83 Appendix E: The Social Economic Database for Florida Counties 85
iii List of Tables Table 4.1 Road Characteristics Database 35 Table 4.2 GIS Layer Metadata 35 Table 5.1 Land Use Reclassification 39 Table 5.2 Definition of Independent Va riables in AADT Prediction Models 43 Table 5.3 Regression Results for Larg e Metropolitan Area, State/County Highway Model 44 Table 5.4 Regression Analysis for Large Me tropolitan Area, Local Street Model 44 Table 5.5 Regression Results for Sma ll-Medium Urban Area, State/County Highway Model 45 Table 5.6 Regression Analysis for SmallMedium Urban Area, Local Street Model 45 Table 5.7 Regression Analysis for Rura l Area, State/County Highway Model 46 Table 5.8 Regression Analysis for Rural Area, Local Street Model 46 Table 5.9 Model Validation for Six Models 48 Table A.1 Descriptive Statistics for Traffic Counts in Hillsborough County 65 Table A.2 Descriptive Statistics for Differe nt Types of Streets in Citrus County 66 Table A.3 Descriptive Statistics for Differe nt Types of Streets in Nassau County 67 Table C.1 Local Street 74 Table C.2 Vehicular Trail 74 Table C.3 Ramp and Circle 74 Table C.4 Other Facility 75
iv Table D.1 County Group based on Population 83 Table E.1 The Social Economic Data base for Florida Counties (2005) 85
v List of Figures Figure 3.1 Suntree Road in Brevard County 26 Figure 3.2 Driveway in Alachua County 26 Figure 3.3 Bismark Road in Nassau County 27 Figure 3.4 Assigning AADT Values to Type I Streets 27 Figure 3.5 County Group based on Population 30 Figure 3.6 Distribution of County Gr oups in the State of Florida 31 Figure 4.1 The Social Economic Databa se for Florida Counties (1995-2005) 34 Figure 5.1 Accessibility to Freeway or State Highway 40 Figure 5.2 Error Distribution of C ounty Highway Testing Counts in Miami-Dade County 49 Figure 5.3 Error Distribution of Lo cal Street Testing Counts in Miami-Dade County 49 Figure 5.4 Spatial Distribution of Erro r Percentage of Te sting Counts in Miami-Dade County 51 Figure 5.5 Error Distribution of County Highway Testing Counts in Citrus County 52 Figure 5.6 Error Distribution of Local Street Te sting Counts in Citrus County 52 Figure 5.7 Spatial Distribution of Error Pe rcentage of Testing Counts in Citrus County 53 Figure 5.8 Error Distribution of Te sting Counts in Sumter County 54 Figure 5.9 Spatial Distribution of Erro r Percentage of Te sting Counts in Sumter County 55 Figure 6.1 The DBF File for Palm Beach County 59
vi Figure B.1 The First Step of Type I Roads AADT Assignment in Flagler County 68 Figure B.2 The Second Step of Type I Roads AADT Assignment in Flagler County 69 Figure B.3 AADT Assignment on Highway 100 70 Figure B.4 AADT Assignment on I-95 71 Figure B.5 AADT Assignment on State Highway 5 72 Figure C.1 A40 Unnamed Street in Brad County (Local Street) 75 Figure C.2 Bismark Road (Local Street) 76 Figure C.3 4wd Road (Vehicular Trail) 76 Figure C.4 Trail (Vehicular Trail) 77 Figure C.5 Trail 2 (V ehicular Trail) 77 Figure C.6 Ramp (Ramp) 78 Figure C.7 Connecting Road (Ramp) 78 Figure C.8 Minnesota Road (Circle) 79 Figure C.9 Suntree Road (Circle) 79 Figure C.10 Lake Andrew (Roundabout) 80 Figure C.11 Diamond Ramp (Ramp) 80 Figure C.12 Service Road (Service Drive) 81 Figure C.13 Driveway 81 Figure C.14 Park Area 82
vii ASSIGNMENT OF ESTIMATED ADVERAGE ANNUAL DAILY TRAFFIC VOLUMES ON ALL ROADS IN FLORIDA Tao Pan ABSTRACT In the first part, this thesis performed a study to compile and compare current procedures or methodologies for the estima tion of traffic volumes on the roads where traffic counts are not easily available. In th e second part, linear regr ession was practiced as an AADT estimation process, which wa s primarily based on known or accepted AADT values on the neighboring state and loca l roadways, population densities and other social/economic data. To develop AADT prediction models fo r estimating AADT values, two different types of database were created, including a social economic database and a roadway characteristics database. Ten years soci al economic data, from 1995 to 2005 were collected for each of the 67 counties in the state of Fl orida, and a social economic database was created by manually imputing da ta obtained from different resources into the social economic database. The roadway characteristics data base was created by joining different GIS data layers to th e Tele Atlas base map provided by Florida Department of Transportation (FDOT). Stepwise regression method was used to se lect variables that will be included into the final models. All selected independent variables in the models are statistically
viii significant with a 90% level of confidence. In total, six li near regression models were built. The adjusted R2 values of the AADT prediction models vary from 0.166 to 0.418. Model validation results show that the MAPE values of the AADT prediction models vary from 31.99% to 159.49%. The model with the lowest MAPE value is found to be the minor state/county highway model for rural area. The model with the highest MAPE value is found to be the local street model fo r large metropolitan area. In general, minor state/county highway models provide more reasonable AADT estimates as compared to the local street model in terms of the lower MAPE values.
1 CHAPTER ONE INTRODUCTION 1.1 Background Information On October 1, 2005, Federal Legislation Â“the Safe, Accountable, Flexible, Efficient Transportation Equity Act: A Leg acy for Users (SAFETEA-LU)Â” created a new Highway safety Improvement Program (HSIP) to reduce traffic fatalities and serious injuries on public roads. In the Section 148(b )(2), states are required to submit an annual report describing on a minimum of 5 percent of the locations with the most hazardous roads. The Federal Highway Administration (FHWA) noted in Title 23, United States Code that the 5 percent report should address locations exhibiting the most severe safety needs on all public roads and the identificati on of hazardous roads should be based on fatalities and serious injuries In an effort to meet th ese SAFETEA-LU requirements the Florida Department of Transportation (FDOT) has purchased a GIS base map from Tele Atlas and is assigning the geogr aphic location of all crashes on all roads for calendar year 2006. It is believed that different measures used may result in different listings of the locations with the most severe safety needs, a mixture of methods may be appropriate. For example, a low volume road that having one or two fatalities or serious injuries in a year may be involved in the most severe list if rate per 100 million vehicle miles traveled
2 (MVMT) is used as the measure, but may not be on the list if rate per mile is used. Conversely, a high volume road like an intersta te highway could possibly have a high severity ranking based on fatal and serious in jury crashes per mile but a relatively low ranking based on such crashes per 100 MVMT. FHWA also required that each state should provide a composite listing based on low volume and high volume roads. As a result, to identify the 5% most hazardous roads, Annual Average Daily Traffic (AADT) information on all roads should be collected and assigned to all roads on the purchased map first. On the other hand, AADT itself is an important measure of crash exposure and also needed to derive other measures like vehicle m iles traveled (VMT). For production of these required analyses by traffic safety engineers, the information of AADT on all roads in Florida is extremely important fo r calculating crash e xposure on every specific roadway segment. In Highway Capacity Manual 2000 (HCM 2000), Annual Average Daily Traffic (AADT) is defined as the total volume passing a point or segmen t of a highway facility in both directions for one year di vided by the number of days in the year. It is one of the most important traffic variables needed for an alysis of traffic cras h rates and is widely used in almost all transportation fields. The state road system a nd many of the major county and local roads have AADT data compiled annually in the DepartmentÂ’s Roadway Characteristics Inventory database. Various offices within the FDOT have tools for estimating traffic volumes on some of the primary local collector and connecting roadways. On state roads or major county or local streets, AADT values are measured by Automatic Traffic Recorders (ATR) installed on the roads. Due to budgetary restraints,
3 AADT counts for some local streets or county roads are often not ava ilable and it is not practical to collect data on the 100,000 miles of local roads in Florida. Sometimes, AADT values on such roads can be estimated by using multiple linear regression models or other transportation demand estimating models Several studies (Q, Xia et al. (1999), F. Zhao et al. (1999) and D. Mohamad et al. (1998)) have developed regression models for estimating the AADT values on off-system roads where traffic counts are not available. For example, in a study conducted by the Flor ida International Univ ersity (FIU) in 1999, four linear regression models were develope d to estimate ADT values on off-system roads for four different area types in Florida. The FIU ADT prediction models include a state-wide model, a rural model, a sm all-medium urban area model and a large metropolitan area model. The R2 values for these models vary from 0.29 to 0.69. Model validation results show that the forecasting er ror of the FIU models varies from 23.73% to 188.00%. All the studies mentioned above mainly fo cused on state maintained roads in state highway system due to the lack of traffic da ta, especially on local streets. Therefore, FDOT proposed this project to develop new methodologies or procedures to estimate AADT on all roads in Florida, and assign the es timated results to the purchased base map. This project was entitled Â“Assignment of Estimated Average Annual daily Traffic Volume on All Roads in FloridaÂ”, and the prin cipal investigator of this project was Dr. Jian Lu, Professor in the Department of Civil and Environmental Engineering at the University of South Florida. This thesis c overed the whole process of the project, which focused on the AADT data collection and pr ocessing, models development and AADT assignment.
4 In this thesis, the author co mpleted the following tasks: 1) Collecting road characteristics data su ch as number of lane, median type, accessibility to freeway, lane use, local e and functional classification for all public roads in Florida, merging and assi gning all these information to the GIS base map given by FDOT, which has roadway characteristics for numerous line segments in all the 67 counties in Florida. This phrase was the most important and time consuming part of the project. 2) Collecting ten years social economic data from 1995 to 2005 for all the 67 counties in Florida, and creating social economic database for the further model development. Seven categories of so cial economic data, including county population, mileage, vehicle, municipality, labor force, income and retail sales, were analyzed in detail. 3) Stepwise regression method was used to se lect variables that will be included into the final models. All selected independent variables in the models are statistically significant with a 90% level of confidence. Totally six linear regression models were built. 4) Model validation was conducted to test whether or not newly created models provided reasonable for all roads in Fl orida. Three counties were randomly selected, and all the 1149 traffic count data within these three counties were considered as testing sites, and not involved in model development. 5) Assigning estimated AADT values to the base map. Given the fact that the various FDOT applications should not be producing or us ing contradictory information, the author provided reasona ble estimations of AADT only for those
5 road segments that did not have reasonable values or estimates from known sources. 1.2 Research Objective and Approaches The main objective of the study is to develop new pr ocedure/methodology for the estimation of traffic volumes on the roads wher e traffic counts are no t easily available, and validate the results with current count data from GIS base map provided by FDOT. This AADT estimation process is primar ily based on known or accepted AADT values on the neighboring state and local road ways, population densities and other social/economic data. More specifically, the research should follow these steps: 1) Identify and compare the existing methods for estimating AADT values on county roads/local streets from any re liable and reput able source. 2) Select the most promising me thod for a) Adoption into this crash analysis process; b) Modification for use in this specific f unction; and/or c) Use in the development of a new estimating model. 3) Adopt or develop models for use in this safety analysis process. 4) Validate the models based on AADT values collected from fields. 5) Select the best model for estimating the AADT values on county/local roads which takes into account the volumes on its neighboring state roads and other social/economic data. 1.3 Organization of the Thesis This thesis consists of five chapters. Chapter 1 provides a brief introduction of the research, including the background of the research, research objectives and past studies conducted in this area. Chapter 2 discusses pa st studies in this field, along with their
6 studyÂ’s key findings and limitations. Chapte r 3 summarizes the methods used for estimating and assigning the AADT values for di fferent types of road in Florida. Issues related with data collection and database s were discussed in Chapter 4. Chapter 5 presents the calibration and validation of AADT prediction models. Finally, Chapter 6 provides a summary and the conc lusions of this research.
7 CHAPTER TWO LITERATURE REVIEW 2.1 Introduction Literature review is conducted to summarize the methods used by previous studies to estimate AADT values on off-syst em roads and to see whether or not the models or results can be used in Florida. Previous studies in this area have mainly focused on two topics, including: (1) the conve rsion of daily traffic volume data obtained from traffic counts into AADT, and (2) th e estimation of AADT values based on regression models. 2.2 Traffic Count The most direct and reliable method for obtaining AADT is to install Automatic Traffic Recorders (ATR) on road segments. Th e ATRs provide continuous traffic volume counts in each day throughout the whole year under ideal conditions. Due to budgetary limitations, it is impractical to install ATRs on all road segments in the State of Florida. As an alternative, Coverage Count is widely used on nonATR road sections for short period traffic count (SPTC). As a short term AADT estimation method, coverage count usually collects 24 to 48 hour traffic volume data in two success days, 48 consecutive hours in rural areas while 24 hours in urban areas to meet the re quirements on minimum count duration in Traffic Monitoring Guide However, it is still labor costly to use coverage count to collect traffic volume on a ll roads in the network, because personnel is
8 needed to install portable traffic count to get data and then turn to the next point. Another short-period traffic count calle d control or seasonal count al so provides continues traffic volume data. Not like coverage count, seasona l count is only used for seasonal factor estimation and seldom mentioned in documents because there are a lot of alternatives available. 2.3 Traffic Count Program Due to limitation of budget, in most st ates, AADT is estimated by multiplying the coverage count data by day of week (DOW) and month of year (MOY) factors, which are estimated from continuous count groups. Even for Coverage count data, because of limitation of personnel resources, some states conduct a traffic count program to collect coverage count data in a three year cycle rath er than collect all the data each year. Annual Growth Factor is applied on those segments without current traffic volume data to calculate AADT data based on hi storical traffi c count data. In Indiana, Indiana department of Transportation (INDOT) sets two Traffic Monitoring Systems to obtain AADT informat ion. 110 ATRs are set on the statewide road network to determine AADT values (the se data are collected 24 hours per day, 365 days per year) as well as seve ral adjustment factors such as Axle, Annual Traffic Growth trends and Seasonal Factor. Besides these 110 permanent traffic c ounters, the statewide coverage count program is applied to coll ect 48 hour traffic count s on all State Highway system segments. A three year program is c onducted to collect the coverage count data, which means one-third of all the segments are counted per year. Average Daily Traffic (ADT) values got from 48 hour traffic covera ge count can be converted into AADT by multiplied with corresponding Seasonal Factor s. Because only one-third of all the
9 segments are counted in each year, AADT of th e rest two/third roads can be calculated from previous data by employing time series model, in which Annual Traffic Growth factor is to be used. Virginia Department of Transportation (VDOT) conducts a tra ffic count program in which 100,000 segments of roads and highways are included. 322 of these segments have traffic count station. Over 250 of the c ontinuous traffic counters are installed on the National Highway System to de termine adjustment factors. In the state of Florida, to meet vari ous transportation need s, a solid Traffic Monitoring Program is operated to estimate AA DT on state maintained roads. Based on more than 6,000 traffic monitoring locatio ns, the estimated AADT on approximately 12,000 miles of state highways can be done, that is, for every two miles of roadway there lies a traffic monitoring (Desai, 2000). In addition to these 6000 traffic monitoring locations, more than 300 season al counters are used to provide continue s traffic data on road network. In doing so, a clear picture is given to display traffi c seasonal pattern, in which seasonal factor is to be calculated to adjust volum e data got discontinuously into AADT. 2.4 AADT Conversion with ATR Data For Automatic Traffic Recorder (ATR) da ta, one simple and precise method to estimate AADT is calculating the mean value of all the 365 daily traffic volume collected in one year. However, in practical terms, ATR data may be insuffi cient or discontinues because of malfunction on recorders, cons truction nearby or data missing. American Association of State Highway and Trans portation Officials (AASHTO) documents describe that most permanent counters retain about 270 of a total of 365 days traffic data,
10 and few permanent counters can keep more than 350 days volume data. As a result, the daily traffic information from device contains some zero data pattern which can be easily found out. These missing data may cause cons iderable bias in AADT calculation. To solve the problem, AASHTO puts forward a sophi sticated algorithm to reduce this kind of bias in Guidelines for Traffic Data Programs In AASHTO method, the first step is to calculate Monthly Average Days of the Week (MADW), which is the average traffic volume for each day of the week for each month. Thus there are seven MADW values (Monday to Sunday) for each month. These calculated results can be marked with MADWi,j where i ( = 1, 2Â… 12) stands for the twelve months in one year a nd j ( = 1, 2Â… 7) represents the seven days in one week. For instance, the average daily traffic on Monday in January can be marked with MADW1,1. To the end of the whole, there are totall y 84 MADW values can be obtained. The second step of the procedure is to calculate the average value of each day across the twelve months, yielding seven Annual Average Days of the Week (AADW) values, and AADT is the arithmetic mean of these seven AADWs, under the potential assumption that the weight of each calendar day is equal. The whole procedure can be expressed as the following equation (1): 12 1 7 1 j i,84 1ijMADW AADT (1) Where i = twelve months in a year ( i = 1, 2Â… 12), j = seven days in a week ( j = 1, 2, Â…, 7), and MADW = monthly average days of the week traffic.
11 2.5 AADT Conversion with Coverage Count Data For coverage count data, methods are also developed to convert the short term traffic counts to AADT. For example, still in AASHTO Guidelines for Traffic Data Programs a standard procedure is provided for c onverting the coverage count data into AADT. The first step of the AASHTO method is to summarize the coverage count data as a one-day, 24 hour traffic volume data. The second step is to multiply the 24-hour axle impulses by the axle correction factor for the presence of vehicles with more than two axles. The preferred approach to obtain th e axle correction factor is to study the continuous data from corresponding grouped vehi cle classification counters for the same days as the short-term traffic count. An aver age number of axles pe r vehicle at permanent counts is calculated based on the 13 vehicle cl assifications used by the Federal Highway Administration (FHWA). A group mean valu e is summed for all similarly grouped counter sites, and the inverse of the group mean is the axle adjustment factor. The third step is to find out the relationship of days of short-term traffic monitoring to the whole year. To remove the difference between traffi c patterns in short peri od count station and that in long term stations, seasonal factors such as the month-of-year (MOY) and day-ofweek (DOW) factors are summarized and cal culated from similar grouped ATR or other continues count station like Control/seas onal Count. The whole procedure currently adopted in Florida for summari zing AADT from short term counts can be presented in the following equation. Axle SF ADT AADT (2)
12 Where Axle = axle correlation factor that converts the counted number of axels to the number of vehicles; ADT = average daily traffic, t ypically the average value of a 72-hour traffic count collected from Tuesday to Thursday; SF = seasonal factor that reflects tr affic seasonal fluctuation pattern; and AADT = estimate of typical daily traffic on a road segment for all days of the week, Sunday though Saturday, ove r the period of one year. The problem with coverage count is that the method still needs the installation of traffic counters to collect traffic volume data Thus, it is impractic al to cover the whole road network in the State of Florida. Instead of using traffic volume data collected from coverage count to estimate AADT, Wang and Teng (2004) used traffic volume data collected by other agencies such as tr affic management centers to estimate AADT. Traffic management center is an important component of intelligent transportation systems (ITS). Traffic volume data collected by traffic management center is originally used for some other purposes such as tr ansportation planning, tr avel time estimation, congestion detection, pavement management, a nd/or air quality analysis. Once used for AADT estimation, the method suffers from a majo r limitation that the ITS data is often not reliable because of insufficient mainte nance work. Wang and Teng compared the ITS data based AADT estimation me thod to the coverage count method. It is observed that with the number of missing da ys increasing, coverage count based AADT is more likely to have smaller errors than ITS data based AADT.
13 Several studies ( AASHTO Guidelines for Traffic Data Programs 1992, S. Gadda et al. (2007) and Traffic Monitoring Guide, (1995)) have also looked at the errors associated with AADT conversion methods. The first type of error described in AASHTO Guidelines for Traffic Data Programs is called sampling error, which is ca used by measure instrument during the procedure of data collection. It is found th at when traffic volume is near 5000 AADT, short period traffic count typi cally like pneumatic road tube can provide results with an error less than 10 percent. When traffic volume reaches 10,000 AADT the error increases to more than 10 percent. This is because axles on several vehicles press on the tube at same time. The short period traffic volume data co llected by coverage counts should be transformed to AADT value by multiplied ad justment coefficients. To obtain these coefficients, there lies an assumption that the traffic pattern in the short-period count site should be equal to that in continues counts. It is not necessarily the same case in real world. As a result, the second type of erro r called factoring erro r occurs. Generally, seasonal factor, axle correcti on factor and Day of Week ad justment factor calculated from continues traffic count like ATR or c ontrol/seasonal count are adopted in such a conversion. Error may increase dramatically if the s hort period volume data is affected by holidays or special events, since these adjustme nt coefficients cannot correct this kind of error. Because these correction factors cannot help re move atypical volume variation caused by holidays or special events.
14 Gadda et al. (2007) conducte d a study to quantify the un certainty related with the AADT estimated using coverage count or other short-term AADT estimation methods. Several error types were menti oned in GaddaÂ’s study, including: 1) Sampling Errors and factoring error 2) Misclassification Error 3) Spatial Error As mentioned above, sampling error is the error generated during the data collection procedure, and factoring error is the error resulting from estimating AADT by using coverage count data. Misclassification er ror occurs when AADT data was assigned to a different site. Spatial error occurs when a road segment is assigned with AADT information obtained from nearby road segments even they are on the same road. In Gadda et al.Â’s study, variati ons in AADT estimation errors were investigated for both Minnesota and FloridaÂ’s ATR sites. It was f ound that including weekend traffic data will result in large errors in th e AADT estimates. In generally, data collected from urban sites suffers from higher error levels as compared to those collected from rura l sites. It is also found that classifying the count sites into different categories based on the functional classification, lane count, and area types w ould help to reduce the estimation error of AADT. 2.6 AADT Estimation Models On roadways where traffic counts are not available, AADT data is often estimated using AADT prediction models. There are tw o major types of AADT prediction models, including time series models and linear regr ession models. Time series models estimate the AADT growth factors based on historical AADT data. The growth factors were used
15 to predict AADT values in fo recasting years. Linear regr ession models estimate the relationships between AADT and various e xplanatory variables. The explanatory variables used in this AADT prediction mode ls often include the ro adway characteristics such as median type, number of lanes, land us e, and/or the functiona l classification of the road, and social-economic variables such as the county population, taxable sales, county lane mileage, and vehicle registration. A study conducted by Mohamad et al. (1998 ) in Indiana developed a multiple linear regression model to estimate the AADT on the county roads where traffic counts are not available. The initially considered independent variables include the following: 1) County Population 2) County Households 3) County Vehicle Registration 4) County Employment 5) County Per Capita Income 6) County Mileage, which includes State Highw ay Mileage, Arterial Mileage, and Collector Mileage. 7) Location: rural or urban 8) Presence of Interstate Highway 9) Accessibility, which is defined as the accessibility to freeways for each road. Stepwise regression method was used to determine which i ndependent variables should be included in the model. Four inde pendent variables were selected. The final AADT prediction model is given as follows:
16 4 3 2 146 0 24 0 84 0 82 0 82 4 X Log X Log X X . AADT Log (3) Where X 1 = Locale (1 = urban; 0 = rural) X 2 = Access (1 = easy access or close to the state highway; 0=otherwise) X 3 = County Population X 4 = Total Arterial Mileage of a county The R2 value of the model is 0.77 which is reasonably high. The major limitation of the study is that the AADT prediction model is developed based on a relatively small database. The model was developed based on 89 traffic counts collected from 40 counties, which means that an average of only 2 traffic counts was availabl e for each county. The most relevant study regarding this t opic was conducted by Zhao et al. in 1999. In that study, 67 counti es in Florida were classified in to three categories based on the population in each county. For each category, a linear regression model was developed for estimating ADT values on off-system road s where traffic counts are not available. The counties with population less than or equa l to 100,000 were defined as rural area. 27 traffic counts obtained from eight rural-area counties were used to build the rural area model. The counties with th e population greater than 10 0,000, but are not located in major metropolitan areas were defined as small-medium ur ban area. 270 traffic counts were randomly selected to develop ADT pr ediction model for the small-medium urban area. Counties located in major metropolitan areas such as the Broward County were defined as large urban area. 443 traffic c ounts were used to develop ADT prediction model for the large urban area. Researchers of that study also de veloped a state-wide
17 model based on 107 county level data obtaine d from 1995 county profile provided by FDOT. In Zhao et al.Â’s study, the independent variables initially considered in the statewide and the rural area models include: 1) Population (POP): the tota l population in a county. 2) Municipality Population (MUNICI): the to tal population in incorporated areas. 3) Labor Force (LABOR): the total labor within a county. 4) Per Capita Income (INCOME): th e per capita income of a county. 5) Taxable Sales (TAXABLE): the ta xable sales of a county. 6) Lane Mile (LANEMILE): the total lane miles of state ro ads in a county. A total of 14 variables were initially c onsidered in the small-medium urban area model development. Th ese variables include: 1) DU_SF: the total single family dwelling units in a Traffic Analysis Zone (TAZ). 2) POP_SF: single family population in a TAZ. 3) SAUTO: total single family automobile ownership in a TAZ. 4) DU_MF: total multi-family dw elling units in a TAZ. 5) POP_MF: multi-family popul ation in the TAZ. 6) MAUTO: total multi-family population in the TAZ. 7) HOT_OCC: population in hotel/motels in a TAZ. 8) IND_EMP: industrial employment in a TAZ. 9) COM_EMP: commercial employment in a TAZ. 10) SER_EMP: Service employment in a TAZ. 11) SCH_ENR: school enrollment in a TAZ.
18 12) LANES: number of lanes at the coun t station location in two directions. 13) ATYPE: area type of the count station location. 14) FTYPE: facility type of the road located the count station. The following variables were initially cons idered in the larg e urban area model: 1) Number of Lane (NUMBEROFLANE): the number of lanes on a roadway. 2) Area Type (AREATYPE): land use types in cludes: Central Business District (CBD), Fringe Area, Residential Area, Ou tlying Business District, and Rural Area. 3) Functional Classification (FCC): state mi nor arterial, count y minor arterial, county collector, city coll ector, local and unclassified. 4) Facility type (FACI): divided arterial, undivided arterial, co llector and centroid collector. 5) Population (POP): the total popu lation within a certain di stance of a count station. 6) Single-family Population (SFPOP): the to tal single-family population within a certain distance of a count station. 7) Single-family dwelling units (SFDUS): th e total occupied single-family housing units within a certain dist ance of a count station. 8) Multi-family dwelling units (MFDUS): th e total occupied multi-family housing units within a certain dist ance of a count station. 9) Auto Ownership (AUTO): the estimated to tal number of automobiles within a certain distance of a count station. 10) Industrial Employment (INDEMP): the total industrial employment number within a certain distan ce of a count station.
19 11) Commercial Employment (COMMEMP): the total commercial employment number within a certain di stance of a count station. 12) Service Employment (SEREMP): the to tal commercial em ployment number within a certain distan ce of a count station. 13) School Enrollment (SEREMP): the total service employment number within a certain distance of a count station. 14) Hotel Occupancy (HTL): the total hotel o ccupants within a certain distance of a count station. 15) Accessibility to State Roads (ACCESS1): th is variable will assume a value of 1 when there are state road s nearby, and 0 otherwise. 16) Accessibility to Off-system Road (ACCESS2): this variable will be 1 when there are other county roads nearby, and 0 otherwise. The final model equations in Zhao et al.Â’s study are given as follows: 1) State-wide model: INCOME POP ADT 1077 0 0057 0 60 9562 (4) Adjusted R2 = 0.1128 2) Rural area model: VEHICLE LANEMILE LABOR POP ADT 003238 0 930235 18 261858 0 122587 0 489444 4853 Adjusted R2 = 0.4488 (5) 3) Small-medium urban model: OCCUPATION COMMERCIAL ATYPE LANES ADT 78 1 85 2 14 1580 23 6770 13418 Adjusted R2 = 0.7206 (6)
20 4) Large metropolitan model: ITY ACCESSIBIL AUTO AREATYPE FCC LANE OF NUMBER ADT 06 1224 15 0 27 1388 57 5227 86 4689 12886 Adjusted R2 = 0.6069 (7) The authors of that study also valida ted the ADT prediction models based on a relatively limited number of traffic counts. Th e mean absolute percen tage errors of the ADT prediction models range from 22.66% to 188.00%. The small-medium urban area model has the best performance in terms of the lowest mean absolute percentage error (22.66%). In summary, Zhao et al.Â’s study provided very useful information about the ADT estimation methods in the State of Florida. Ho wever, the models deve loped in that study cannot be directly used in our project becau se of the following two reasons: (1) Zhao et al.Â’s study was focused on estimating the ADT of off-system roads in Florida while the objective of our study is to estimate AADT va lues of off-system roads; and (2) The models in Zhao et al.Â’s study were developed and validated based on a limited number of traffic counts. It is generally believed that the forecasting capability of AADT prediction model will increase if it is based on a la rge sample of traffic volume counts. In a study conducted by J.K. EOM et al. (2006) in the North Carolina State University, a spatial regression model was developed for estimating the AADT values on county roads where traffic count s were often not available. It was the first time that AADT was estimated from a spatial regression model, which takes into account the spatial correlation between AADT at one locatio n and those at its ne ighboring locations. The thinking behind this method is that tr affic volume at one monitoring station is correlated with the volu me at its neighboring stations. 200 traffic counts were selected
21 randomly out of all the 1154 available counts in Wake County, Nort h Carolina. In the process of sampling data, traffic counts on freeways like I-40, I-440, and US-1 were excluded from the entire database because th e high percentage of through traffic on freeway hurts the spatial rela tionship with traffic volume on surrounding roads. It was found that spatial regression models provided better AADT estimates as compared to ordinary linear regression models if spatia l correlation between AADT at one location and those at its neighbor exists. However, th e conclusion needs further validation because of the small sample size and ignorance of freeways. Since only 200 samples were selected in the study for model developing, th ere lies a question th at how representative these sample stations are and whether or not the model is biased towards the selected samples. 2.7 Other AADT Estimation Methods Tang et al. (2003) used historical and current-year volume data from Hong Kong core traffic count station to compare four different forecas ting models for traffic flow estimation. The four models included: 1) Autoregressive Integrated moving Average (ARIMA) Model 2) Neural Network Model (NN) 3) Nonparametric Regression (NPR) 4) Gaussian Maximum Likelihood (GML) Model ARIMA model is used to forecast both non-seasonal and seasonal data an can only be applied to stationary time series. Neur al network model (NN) applies the idea of writing software based on the structure of th e human brain and consist of many simple processing elements called neurons. Nonparametr ic regression (NPR) models perform in
22 a sense that is more dynamic than the ti me series and neural network models. Nonparametric regression performs predic tion based on a group of similar past cases defined around the current input state at th e time of prediction. GM L models explicitly make use of historical traffic information and real time information in an integrated way The two key random variables considered in the GML model were flow and flow increments with a time interval of five minutes. In that research, data within a period from January 1994 to December 1998 was chosen as historical data for the model development while Janua ry to December 1999 was chosen as current-year for the model validation. Two m easures of performance, the mean absolute error (MAE) and the mean square error (MSE), were selected for comparing the results of the four models. It was pointed out that the ARIMA a nd NN models require extensive data calibration, but the NPR and GML models do not require data calibration and can be implemented easily. The GML model was found to be more promising and robust for extensive applicati on in AADT estimation. M. McCord et al (2002) conducted a project to estimate AADT information by analyzing high resolution sate llite imagery. However, it seemed not an easy task to achieve because the Â“noiseÂ” associated with inferring average traffic conditions from satellite imagery should be small enough and the quantity of images should be great enough that the information can be combined with ground-based data to improve estimation performance. Although the result given in that project s how that high resolu tion satellite imagebased estimation method works as reinfo rcement of ground-based AADT estimation,
23 satellite image analysis still cannot be used widely because of the cost of obtaining and processing image data. 2.8 Summary Traffic volume collection strategies curre ntly adopted cannot cover all the road segments in the whole network, especially li mited data resources are available for off system roads. Although some attempts have been made to set up models or procedures of AADT estimation in the past few years, most of them focused on state highways due to the limitation of traffic data. Results or mode ls in those past studies seemed not suitable for this research, and new methodologies or models are needed for the AADT estimation on all roads in the stat e of Florida. Grounded on the achie vements of previous researches, multiple linear regression is proved to be a promising and dependable method to estimate AADT, which is strict in met hodology and easy in practice.
24 CHAPTER THREE METHDOLOGY 3.1 Introduction The objective of this study is to develop a method/procedure for estimating the AADT values on off-system roads where traffi c counts are not available, and validate the estimated results with current count data fr om the Tele Atlas digital map provided by the FDOT. The Tele Atlas digital map is a GIS ba sed map which contains almost all roadway segments in the State of Florida. Each roadway segment in the digital map is assigned with a unique variable called Dynamap_ID. The FDOT also provided an AADT data base which included the AADT values for about 2.35% roadway segments in the State of Florida. The AADT database was joined to the Tele Atlas base map based on the same Dynamap_ID of each road segment. The Tele Atlas digital map also provides th e functional classification codes for the roadways. In order to achieve the research objectives, the streets provided by the Tele Atlas digital map were divided into three different types based on the number of traffic counts available to each street as well as the functi onal classification codes provided by the base map. The descriptive statistic s for traffic counts in Hills borough County, Citrus County and Nassau County were given in Table A-1 through A-3. The Type I streets include all freeways and major state highways where each ro ad has at least one traffic count in each
25 county. In total, the Type I streets account fo r about 10-15% of the streets in the Tele Atlas digital map. Given the f act that the various FDOT applications should not be producing or using contradictor y information the author need to provide reasonable estimations of AADT only for those road segm ents that do not have reasonable values or estimates from known sources. Due to this reason, AADT values on Type I streets should not be estimated using AADT prediction models because each Type I street has at least one traffic count in each county. In this study, AADT values for Type I streets were assigned manually. Each roadway segment was assigned with the traffic counts collected from the closest roadway segment. The met hod used for assigning AADT values to Type I streets was described in the next subsection. The Type II streets include minor state and county highways, and local streets. Less than 10% of these streets have traffic count s. In total, the Type II streets account for about 80-85% of the streets in the Tele Atlas digital map. AADT values on Type II streets were estimated based on AADT pred iction models developed in this study. About 5% of the streets were defined as Type III streets. The Type III streets include: vehicle trails, freeway ramps, culde-sacs, traffic circles, service drives, driveways, roads in parking area, and alleys. Traffic c ounts on Type III streets are extremely limited, and the samples available to this study are too small to build an AADT prediction model. Due to this reason, we feel that it is very hard to estimate the AADT values on these streets wit hout large-scale field data co llection. Several aero photos of Type III streets are given in Figure 3.1 through 3.3.
26 Figure 3.1: Suntree Road in Brevard County Figure 3.2: Driveway in Alachua County
27 Figure 3.3: Bismark Ro ad in Nassau County Figure 3.4: Assigning AADT Valu es to Type I Streets I-4 I-275 Count 1 Count 2 Count 3 Count 4 Segment A I-4 I-275
28 3.2 Methods for Assigning AADT Values to Type I Streets Type I streets include freeways and majo r state highways where sufficient traffic counts are available. The task for AADT a ssignment on Type I streets is to assign the traffic counts obtained from traffi c count locations to all the segments on the same street. When assigning the AADT values on Type I streets, the followi ng principles were followed: (1) Traffic count obtained from a pa rticular road segment was only assigned to the roadway segments on the same street; and (2 ) If a street in a county has more than one traffic count station, road segments on the same street were assigned with the AADT values obtained from the nearest traffic c ount station. The logic is illustrated by the example given in Figure 3.4. The purpose of th e example is to assign AADT value to the segment A. AADT value collected from traffic count station 2 and 3 will not be considered because they are not on the same street with segment A. Both count station 1 and 4 are on the same street with segment Â“AÂ”. In this case, AADT value collected from count station 1 will be assigned to segment Â“A Â” because it is the nearest traffic count station on the same street. In this study, AADT values for Type I streets were assigned manually. It is extremely time consuming and labor intensiv e to do so for 67 counties in the State of Florida. Using Flagler County as an example, the general procedures for assigning AADT values for Type I streets were illustrated in Figure B-1 through B-5 and briefly described as follows: 1) Step 1: In the Tele Atlas digital map, sele ct the Type I streets and traffic counts on these streets. 2) Step 2: Separate the Type I streets from other streets in the digital map.
29 3) Step 3: There are three Type I streets in Flagler County. Each street was selected by attribute query based on the street name. The selected street was exported into a new shape file. Traffic counts on the sa me street were also selected by using spatial query method provided by ArcGIS. 4) Step 4: Assign the AADT values to all s ections on the same street based on the spatial distance. 3.3 Methods for Assigning AADT Values to Type II Streets Previous studies have demonstrated th at linear regression models can provide reasonable AADT estimates for off-system ro ads where traffic counts are not available (Q, Xia (1999), F. Zhao (1999) and D. Mohamad (1998)). In this study, multiple linear regression models were developed for estima ting AADT values for Type II streets. The linear regression model takes on the following functional form: j jX X AADT ...1 1 0 (8) Where AADT = the dependent variable; Xi = the value of ith independent variable, i=1, 2, 3 Â…n; 0 = constant term; j = regression coefficient for the ith independent variable; = error term; n = number of independent variables. In this study, 67 counties in the State of Fl orida were divided into three area types based on the county population in 2005. The area types considered in this study include:
30 1) Large Metropolitan Area Group ( population > 400,000 ) 2) Small-Medium Urban Area Group ( 100,000 < population < 400,000 ) 3) Rural Area Group ( population < 100,000 ) Figure 3.5 present the county grouping info rmation as well as the population data in each county. 12 counties were included in the large metropolitan area group. Population in these counties accounts for 67.4 3% of total population in the State of Florida. The small-medium urban group includes 22 counties. The population in these counties accounts for about 26.40% of total population in the Stat e of Florida. A total of 33 counties were included in the rural ar ea group and the total population in these counties accounts for 6.17% of total population in the State of Florida. The spatial distribution of the county groups is given in Figure 3.6. Figure 3.5: County Group based on Population Population and County Number Compariason between Groups 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00% Large MetropalitanSmall-Me dium UrbanRural AreaPercentage Population Percentage County Percentage
31 Figure 3.6: Distribution of County Groups in th e State of Florida
32 CHAPTER FOUR DATA COLLECTION 4.1 Introduction The major purpose of data collection work in this study was to collect data used for developing AADT prediction models. Th e AADT prediction models were used for estimating AADT values on Type II streets wh ich, as mentioned above, account for about 80-85% of the streets in the Tele Atlas digital map. Extensive data collection is conducted to cover most possible potential factors that have significant impacts on AADT and great efforts are also made to co mpile and process these data. Two different types of data were collected from different resources, including social-economic data and roadway characteristics data. 4.2 Roadway Characteristics Database Most of the roadway characteristics inform ation used in this study is provided by the digital maps provided by the Tele Atlas. Tele Atlas provides a GIS based digital map in which roadway networks are composed of line segments. The base map includes almost all street segments in the State of Florida. More specifically, the information provided by the base map includes: 1) Dynamap_ID: it is the key variable that was used as a unique identification of each road segment and traffic count in the whole street network; 2) Name: names of road segments. Segments on the same road share the same name.
33 3) FCC (Feature Classification C odes): it is a very important variable which defines the functional type of roadways. A roadway characteristics database was cr eated by joining different data resources to the vector based, geography base map provi ded by the Tele Atlas. An example of the roadway characteristics database is given in Table 4.1. Most of the GIS data layers were obtained from the FDOT websit e except the land use data la yer, which was obtained from the website of the Florida Geographic Data Library (FGDL). More specifically, the GIS data layers which need to be joined to the Tele Atlas digital map include: 1) Urban/rural data layer; 2) Number of lanes data layer; and 3) Land use data layer. These GIS data layers were joined to the Tele Atlas digital map based on their spatial relationships. It is important to note that the data collection work in this study is very time consuming and labor intensive because each data layer has a different geographic coordinate system and cannot be dir ectly joined to the Te le Atlas digital map. Table 4.2 describes the coordination system of each GIS data layer. The author have developed procedures to join different data layers to the Tele Atlas digital map using ArcGIS but it is very time consuming to do so for 67 counties. The most difficult part the author found was to join the land use data la yer to the base map. It takes about 20 hours for the computer to join the land use da ta to the digital map for the county like Hillsborough.
34 4.3 Social Economic Database Ten years social economic data, from 1995 to 2005 was collected for each of the 67 counties in the State of Florida. Social economic data was collected from different data resources such as the website of stat e and county governments and the US census bureau. Social economic data in some years wa s not available. Social economic data in these years was extrapolated from the data in other years. A soci al economic database was then created by manually imputing data obt ained from different resources into the social economic database. A picture of the so cial economic database is given in Figure 4.1. The social economic database includes ag gregated data in county level including county population, total lane mileage, vehicle registratio n, municipality, labor force, average income, and retail sales. Figure 4.1: The Social Economic Data base for Florida Counties (1995-2005)
35 Table 4.1: Road Characteristics Database Dynamap_ID Street Name FCC Land Use Road Side Locale Lane Count Access 0_5mile Access 1mile Access 1_5mile 386,914,608 CR581 A35 12 R 1 2 0 1 1 386,914,608 Bruce B Downs A35 12 R 1 2 0 1 1 386,914,633 CR581 A35 12 R 1 2 0 1 1 386,914,633 Bruce B Downs A35 12 R 1 2 0 1 1 386,914,636 County Hwy 581 A35 12 R 1 2 0 1 1 386,914,636 Bruce B Downs A35 12 R 1 2 0 1 1 386,914,684 Veterans A15 8 L 1 2 0 0 0 386,914,712 Debuel A41 2 1 0 0 0 386,914,718 I-75 A15 13 1 0 0 0 386,914,723 County Hwy 685A A41 2 C 1 2 0 0 1 386,914,723 Van Dyke A41 2 C 1 2 0 0 1 386,914,724 CR685A A41 11 1 0 0 0 386,914,724 Simmons A41 11 1 0 0 0 386,914,725 Juanita A41 11 1 0 0 0 386,914,727 County Hwy 685A A41 11 L 1 1 0 0 0 386,914,727 Van Dyke A41 11 L 1 1 0 0 0 386,914,737 Veterans A15 8 L 1 2 0 0 0 386,914,741 Cypress A41 2 1 0 0 0 386,914,744 Debuel A41 1 L 1 4 0 0 0 386,914,746 Debuel A41 12 L 1 4 0 0 0 386,914,753 Cypress A41 11 1 0 0 0 Table 4.2: GIS Layer Metadata Data Layer Geometry Type XY Coordinate System Datum Units Street Line Lat Long WGS84 D_WGS_1984 Degree Traffic Count Point GCS_WGS_1984 D_WGS_1984 Degree Urban Boundary Polygon NAD_1983_UTM_ZONE_17N D_No rth_American_1983 Meter Number of Lane Line NAD_1983_UTM_ZONE_17N D_North_American_1983 Meter Land Use Polygon Albers Conical Equal Area (Florida Geographic Data Library) D_North_American_1983_HARN Meter County Boundary Polygon Lat Long WGS84 D_WGS_1984 Degree
36 CHAPTER FIVE MODEL CALIBRATION AND VALIDATION 5.1 Model Calibration 5.1.1 Introduction In this study, the counties in Florida we re divided into three groups based on the population in each county. The counties w ith the population less than 100,000 were classified into the rural area group. The counties with the population between 100,000 and 400,000 were classified into the smallmedium urban area group. The counties with the population greater than 400,000 we re classified into the la rge metropolitan area group. In each group, two models were developed for estimating the AADT values on Type II streets, including a state/count y highway model and a local street model. The dependent variable of the state/county highway model is the AADT values on minor state/county highways. In Tele Atlas base map, these road s have the functional cl assification codes of A3X. The dependent variable of the local street model is the AADT values on local streets which have the functiona l classification codes of A4X. 5.1.2 Variable Description The dependent variable of the AADT m odel is the AADT value on a particular street segment. The initially considered inde pendent variables are gr ouped into two types, social economic variable and roadway characteristics variable.
37 There are totally seven initial social economic variables included in the model development. 1) Population. The tota l population in a county. Populat ion is taken as independent variable based on the assumption that population within one area have a significant impact on traffic volume. 2) Total Lane Mileage of Highways. The Tota l lane mileage of hi ghways in a county. 3) Vehicle Registration. The total number of re gistered vehicles in one county. There lies an assumption that the more vehicles registered in a county, the more traffic volume will be loaded on the roadway network in the same county. 4) Personal Income. The per capita income of a county. Accounting to trip generation theory, daily traffic will increase with personal income. 5) Retail Sales. Yearly retail sales in each county. Similar to personal income, it is believed that daily traffic increase with the development of social economy. 6) Municipalities. Population w ithin incorporated area. 7) Labor Force. Labor force within one count y. It is reasonable that more labor attracts more traffic volume within a county. In addition of these seven so cial economic variables, ther e are five initial roadway characteristics variables include d in the model development. 1) Divided/not. A binary vari able to indicate the type of median: divided or undivided. 2) Number of lanes in both sides. The to tal number of lanes in both sides of roadways.
38 3) Location (rural or urban). A binary variable indicating the type of location: urban or rural. The variable is assumed to have a significant relationship with AADT. Roads within urban areas will carry more daily traffic as compared to those within rural areas. 4) Land use. The abutting land use type of a ro ad segment. It is believed that there lies a strong relationship between AADT and land use, with which volume distribution varies dramatically. In th is study, land use data was originally collected from the FGDL website as GIS shape files and joined to the Tele Atlas GIS base map based on the spatial relati onship. The original land use shape files provided by FGDL contain 15 land use t ypes. They were reclassified into 8 similar categories, including public-semipublic, agriculture, commercial, institutional, residential, recreation, industrial and others. The reclassification of land use data is described in Table 5.1. Ei ght binary variables are defined for the eight land use types. 5) Accessibility to Freeways. Unlike other va riables, accessibility will be added into roadway characteristics database as a new variable. It is adapted to judge whether roadway segments fall into areas aff ected by freeways or major state highway, that is, Type I roads. Based on literature review and small sample tests, three buffer sizes are finally selected. They ar e 0.5 mile, 1 mile and 1.5 miles. It means road segments fall in a distance of 0.5 mile 1 mile and 1.5 mile from freeways or state highways will be highlighted and marked separately. Figure 5.1 is the sketch map of the three buffer ranges, in which the central heavy line stands for State Highway 20, shaded pattern for 0.5 mile bu ffer area, striped area for 1.0 mile and
39 dotted area for 1.5 mile. Road segments w ithin different buffer areas are marked with different colors. Table 5.1: Land Use Reclassification Land Use Description Land Use Description PublicSemipublic Public Schools Public Hospital Gov. Owned Leased by NonGov. Lessee Utilities Industrial Manufacturing Lumber Yard Fruit, Meat Packing Canneries Warehouse Industrial Storage Commercial Stores Shops Office Supermarket Shopping Malls and Centers Airports, Marinas and Bus Terminals Restaurants Financial Institution Theater and Stadium Night Club and Bar Hotel and Motel Agriculture Timberland Cropland Grazing Land Institutional Churches Private School Private Hospital Colleges Other Mining and Gas Rivers and lakes Undefined Residential Family Mobile Homes Condo Recreation Forest, Park Golf
40 Figure 5.1: Accessibility to Freeway or State Highway
41 5.2 Model Development The dependent variable of the AADT m odel is the AADT value on a particular street segment. A total of 26721 traffic count s provided by the FDOT were used to build six AADT prediction models. The initially c onsidered independent variables include seven social economic variables and fourteen independent variables. The definition of independent variables is given in Table 5.2. Stepwise regression method was used to select variables that will be included into the final models. In total, six linear regression models were built. The regression results of the AADT predic tion models were gi ven in Table 5.3 through 5.8. The final equations of the AADT prediction models and the adjusted-R2 values of the models were given as follows: Large Metropolitan Area, State/County Highway Model: (9) 186 0 R SEMIPUBLIC 47 585 L RESIDENTIA 648 782 MILE 5 0 601 796 INCOME 069 129 NAL INSTITUTIO 231 1311 NE NUMBEROFLA 252 421 RE AGRICULYTU 185 2839 LABORFORCE 845 8 LOCATION 677 6259 COMMERCIAL 442 2983 DIVIDED 347 1273 VEHICLE 541 13 8 848 AADT2 adj Large Metropolitan Area, Local Stre et Model: (10) 242 0 R VEHICLE 345 4 POPULATION 369 17 LABORFORCE 545 19 COMMERCIAL 194 769 SEMIPUBLIC 226 1040 NE MUNBEROFLA 492 259 LOCATION 195 2745 MILE 5 1 182 567 L RESIDENTIA 459 452 DIVIDED 659 1349 TIES MUNICIPALI 806 3 443 2738 AADT2 adj
42 Small-Medium Urban Area, State/Count y Highway Model: (11) 259 0 R INDUSTRIAL 666 1072 MILEAGE 43 0 SEMIPUBLIC 103 765 L RESIDENTIA 282 431 MILE 5 1 963 952 TIES MUNICIPALI 311 13 SALES 994 0 POPULATION 70.869 VEHICLE 673 27 NE NUMBEROFLA 82 960 COMMERCIAL 767 2760 LABORFORCE 079 122 LOCATION 145 5566 374 770 AADT2 adj Small-Medium Urban Area, Local Street Model: (12) 166 0 R RECREATION 814 2011 NAL INSTITUTIO 1464.231 COMMERICAL 556 1491 INDUSTRIAL 091 3320 TIES MUNICIPALI 9437 0 POPULATION 468 14 VEHICLE 468 18 LOCATION 119 2707 MILE 5 1 874 2107 L RESIDENTIA 405 679 DIVIDED 69 2482 94 1533 AADT2 adj Rural Area, State/C ounty Highway Model: (13) 378 0 R L RESIDENTIA 708 748 POPULATION 239 33 INDUSTRIAL 493 2324 RECEATION 919 3312 SALES 931 1 LABORFORCE 293 22 E AGRICULTUR 733 1656 TIES MUNICIPALI 072 57 VEHICLE 722 17 LOCATION 551 3878 747 3015 AADT2 adj Rural Area, Local Street Model: (14) 418 0 R L RESIDENTIA 873 1017 E AGRICULTUR 085 1445 LOCATION 501 1458 POPULATION 168 62 505 1225 AADT2 adj The adjusted R2 values of the models vary from 0.186 to 0.418. The R2 values are not un-acceptable considering the fact that the AADT prediction models are, in fact, disaggregate models for which the dependent variables are AADT values for a particular road segment. All independent variables are st atistically significantly with a 90% level of confidence. The author also tested possi ble multicollinearity between independent
43 variables. It was found that the multicollin earity does exit between several independent variables and, as a result, some of the coefficients in the model do not have the expected signs. These correlated independent variables were still included in the models because: (1) later conducted model validation work sh ows that keeping these variables in the model helps reducing prediction errors; and (2) the objective of AADT models are to estimate the AADT values, not to identify the impacts of various independent variables. Table 5.2: Definition of Independent Variables in AADT Prediction Models Social economic Variables Population = population in thousands Mileage = total mileage of highways in a county Vehicle Registration = the total number of registered vehicles in thousands Personal Income = the per capita income in thousands Retail Sales = yearly retail sales in million Municipalities = population within incorporated area in million Labor Force = labor force with in one county in thousands Road Characteristics Variables Variable Name Assigned Value Divided/not Divided = 1, and 0 otherwise Number of lane Number of lanes in both directions Location Urban = 1, and 0 otherwise 0.5 Mile Roads within 0.5 mile from freeways = 1, and 0 otherwise 1.0 Mile Roads within 1.0 mile from freeways = 1, and 0 otherwise 1.5 Mile Roads within 1.5 mile from freeways = 1, and 0 otherwise Public-Semipublic Land use type is Public-Semipublic =1, and 0 otherwise Commercial Land use type is Commercial =1, and 0 otherwise Agriculture Land use type is Ag riculture =1, and 0 otherwise Institutional Land use type is In stitutional =1, and 0 otherwise Residential Land use type is Re sidential =1, and 0 otherwise Recreation Land use type is R ecreation =1, and 0 otherwise Industrial Land use type is I ndustrial =1, and 0 otherwise
44 Table 5.3: Regression Results for La rge Metropolitan Area, State/County Highway Model Parameters Coefficients Standard Error t-statistic Significance Level Constant -848.800 766.550 -1.107 0.268 Vehicle 13.541 0.443 30.572 0.000 Divided 1273.347 204.053 6.240 0.000 Commercial 2983.442 227.294 13.126 0.000 Location 6259.677 529.653 11.818 0.000 Laborforce -8.845 0.716 -12.355 0.000 Agriculture -2839.185 389.819 -7.283 0.000 Numberoflane 421.252 69.617 6.051 0.000 Institutional 1311.231 383.175 3.422 0.001 Income 129.069 26.513 4.868 0.000 0_5mile 796.601 196.480 4.054 0.000 Residential -782.648 248.232 -3.153 0.002 Semipublic -585.470 279.777 -2.093 0.036 R2 = 0.186, R2 adj = 0.186 Table 5.4: Regression Analysis for Large Metropolitan Area, Local Street Model Parameters Coefficients Standard Error t-statistic Significance Level Constant -2738.443 437.939 -6.253 0.000 Municipalities 3.806 0.726 5.238 0.000 Divided 1349.659 212.907 6.339 0.000 residential -452.459 183.301 -2.468 0.014 1_5mile -567.182 184.731 -3.070 0.002 location 2745.195 393.557 6.975 0.000 Numberoflane 249.492 86.614 2.880 0.004 semipublic 1040.226 249.959 4.162 0.000 Commercial 769.194 218.337 3.523 0.000 Laborforce -19.545 1.238 -15.782 0.000 Population 17.369 1.055 16.457 0.000 Vehicle -4.345 0.816 -5.323 0.000 R2 = 0.244, R2 adj = 0.242
45 Table 5.5: Regression Results for Sm all-Medium Urban Area, State/County Highway Model Parameters Coefficients Standard Error t-statistic Significance Level Constant 770.374 404.301 1.905 0.057 location 5566.145 247.125 22.524 0.000 Laborforce 122.079 7.972 15.313 0.000 Commercial 2760.767 207.855 13.282 0.000 Numberoflane 960.820 88.258 10.887 0.000 Vehicle 27.673 1.831 15.114 0.000 Population -70.896 4.366 -16.237 0.000 Sales 0.994 0.195 5.107 0.000 Municipalities -13.311 2.365 -5.628 0.000 1_5mile 952.963 196.098 4.860 0.000 Residential -431.282 219.761 -1.963 0.050 Semipublic 765.103 288.482 2.652 0.008 Mileage -0.430 0.186 -2.309 0.021 Industrial 1072.666 508.713 2.109 0.035 R2 = 0.261, R2 adj = 0.259 Table 5.6: Regression Analysis for SmallMedium Urban Area, Local Street Model Parameters Coefficients Standard Error t-statistic Significance Level Constant 1533.940 647.179 2.370 0.018 Divided 2482.690 350.562 7.082 0.000 Residential -679.405 294.645 -2.306 0.021 1_5mile 2107.874 337.002 6.255 0.000 Location 2707.119 476.108 5.686 0.000 Vehicle2 18.468 2.430 7.600 0.000 Population -14.468 2.597 -5.570 0.000 Municipalities 9.437 3.141 3.004 0.003 Industrial 3320.091 919.746 3.610 0.000 Commercial 1491.556 379.700 3.928 0.000 Institutional 1464.231 585.513 2.501 0.012 Recreation 2011.814 828.079 2.429 0.015 R2 = 0.172, R2adj = 0.166
46 Table 5.7: Regression Analysis for Ru ral Area, State/C ounty Highway Model Parameters CoefficientsStandard Errort-statisticSignificance Level Constant 3015.747 249.065 12.108 0.000 Location 3878.551 262.420 14.780 0.000 Vehicle 17.722 11.007 1.610 0.108 Municipalities 57.072 14.166 4.029 0.000 Agriculture -1656.733 224.269 -7.387 0.000 Laborforce 22.293 6.018 3.704 0.000 Sales -1.931 0.886 -2.180 0.029 Recreation -3312.919 712.132 -4.652 0.000 Industrial -2324.493 822.165 -2.827 0.005 Population 33.239 14.270 2.329 0.020 Residential -748.708 267.852 -2.795 0.005 R2 = 0.382, R2adj = 0.378 Table 5.8: Regression Analysis fo r Rural Area, Local Street Model Parameters Coefficients Standard Error t-statistic Significance Level Constant 1225.505 384.195 3.190 0.002 Population 62.168 9.365 6.639 0.000 Location 1458.501 503.887 2.894 0.004 Agriculture -1445.085 483.470 -2.989 0.003 Residential -1017.873 471.691 -2.158 0.032 R2 = 0.432, R2adj = 0.418 5.3 Model Validation The purpose of model validation is to test if the developed AADT prediction models can provide reasonable AADT estimates fo r Type II streets in the State of Florida. Traffic counts from three randomly selected counties were used for validating AADT prediction models. These traffi c counts were not used for m odel calibration described in
47 the previous section. The Mean Absolute Per centage Error (MAPE) is used to evaluate the forecasting capability of the AADT pred iction models. The MAPE value measures the prediction error between the AADT values estimated using AADT prediction models and those obtained from traffic count sta tions. The definition of MAPE is given as follows: n 1 i Fi i F i MAADT AADT AADT n 1 MAPE. (15) Where AADTFi = the ith field measured AADT value, i=1, 2, 3 Â…n; AADTMi = the ith AADT value estimated by AADT prediction model, i=1, 2 Â…n; n = sample size In total, 1149 traffic counts from thr ee counties were used for AADT model validation. Model validation results are give n in Table 4.9. The MAPE values for six AADT prediction models vary from 31.99% to 159.49%. The model with the lowest MAPE value is found to be the minor state/county highway model for rural area. The model with the highest MAPE value is found to be the lo cal street model for large metropolitan area. In general, minor stat e/county highway models provide more reasonable AADT estimates as compared to the local street model in terms of the lower MAPE values. In this study, the local streets we re defined as the Type II streets with FFC of A4X. It is not a surpri se that local street models provide relatively poor AADT estimates since these roads have much fewer traffic counts available as compared to minor state/county highways.
485.3.1 Model Validation for Large Metropolitan Area The frequency distributions of MAPE values for the large metropolitan area models are given in Figure 5.2 and 5.3. Th e models were tested against the AADT data collected in Miami-Dade County. As shown in Figure 5.2 and 5.3, the vast majority of the MAPE values for both minor state/count y highway model and lo cal street model are less than 50%. The spatial distribution of forecasting errors in Miami-Dade County is also given in Figure 5.4. Table 5.9: Model Vali dation for Six Models County Group Functional Classification N MAPE Min Max Standard Deviation Large Metropolitan (Miami-Dade County) County Highway 870 46.81% 12.90%809.30% 0.664 Local Street 123 159.49%2.51% 974.72% 1.820 Small-Medium Urban (Citrus County) County Highway 112 65.01% 1.05% 609.88% 0.963 Local Street 20 65.35% 3.36% 213.24% 0.569 Rural (Sumter County) County Highway 22 31.99% 0.19% 93.87% 0.252 Local Street 2 46.79% 46.27%47.325% 0.007
49 Figure 5.2: Error Distributi on of County Highway Testing Counts in Miami-Dade County Figure 5.3: Error Distributi on of Local Street Testing Counts in Miami-Dade County Error Distribution0 100 200 300 400 500 600 700 <50%50%-100%100%-150%150%-300%>300%Error in PercentageFrequency Error Distribution0 5 10 15 20 25 30 35 40 45 50<50%50%-100%100%-150%150%-200%200%-300%300%-500%>500%Error in PercentageFrequency
505.3.2 Model Validation for Small-Medium Urban Group The frequency distributions of MAPE va lues for the small-medium area models are given in Figure 5.5 and 5.6. The models were tested against the AADT data collected in Citrus County. As shown in Figure 5.5 and 5.6, the vast majority of the MAPE values for both minor state/county hi ghway model and local street model are less than 100%. The spatial distribution of forecas ting errors in Citrus County is also given in Figure 5.7. 5.3.3 Model Validation for Rural Area Group A limited number of traffic counts were provided by the Sumter County. The data was used to validate the AADT prediction models for rural area. The frequency distribution of MAPE values for the minor state/county high way model is given in Figure 5.8. The frequency distributio n of MAPE values for the lo cal street model cannot be developed because the number of traffic count s is too few. As shown in Figure 5.8, the vast majority of the MAPE values for minor state/county highway mo del is rural area is less than 50%. The spatial dist ribution of forecasting errors in Sumter County is given in Figure 5.9.
51 Figure 5.4: Spatial Distribution of Error Percentage of Te sting Counts in MiamiDade County
52 Figure 5.5: Error Distributi on of County Highway Testing Counts in Citrus County Figure 5.6: Error Distributi on of Local Street Testing Counts in Citrus County Error Distribution0 10 20 30 40 50 60 70 <50%50%-100%>100%Error in PercentageFrequency Error Distribution0 1 2 3 4 5 6 7 8 9 10 <50%50%-100%100%-150%150%-200%200%-250%Error in PercentageFrequency
53 Figure 5.7: Spatial Distribution of Error Perc entage of Testing Count s in Citrus County
54 Figure 5.8: Error Distribution of Testing Counts in Sumter County Error Distribution 0 1 2 3 4 5 60%-10%10%-20%20%-30%30%-40%40%-50%50%-60%60%-70%70%-80%80%-90%90%-100%Error in PercentageFrequency
55 Figure 5.9: Spatial Distribution of Error Perc entage of Testing C ounts in Sumter County
56CHAPTER SIX SUMMARY AND FINAL RESULT 6.1 Summary The main objective of the study is to develop new pr ocedure/methodology for the estimation of traffic volumes on the roads wher e traffic counts are not easily available. This AADT estimation process is primarily based on two categories of data. One is known or accepted AADT values on the neighbor ing state and local roadways, ArcGIS was applied to merge and create road characteri stics database from va rious data resources; the other type of data is so cial/economic data like population densities, total lane mile and retail sales. To achieve the research object ives of this study, the st reet segments provided by the Tele Atlas GIS base map were divided into three diffe rent types based on the number of traffic counts available to each street. The Type I streets include all freeways and major state highways where each road has at l east one traffic count in each county. That means there are sufficient traffic counts avai lable on Type I roads and AADT values on Type I streets were assigned manually by assigning AADT values measured from several traffic count stations to all other segments of the same ro ad. In total, the Type I streets account for about 10-15% of the street s in the Tele A tlas base map. The Type II streets include minor state and county highways and local streets. Less than 10% of these streets have traffic counts available. AADT values on Type II
57 streets were estimated based on six linear regr ession models developed in this study. In total, the Type II streets account for about 8085% of the streets in the Tele Atlas base map. About 5% of the streets were defined as Types III streets. The Type III streets include vehicle trails, freeway ramps, cul-de-sac, traffic circles, serve drives, driveways, roads in parking area, and alleys. Traffic counts on these Type III streets are extremely limited, and the samples available to this st udy are too small to build an AADT prediction model. Due to this reason, we feel that it is very hard to estimate the AADT values on these streets without large-sc ale field data collection. To develop AADT prediction models for estimating AADT values on Type II streets, two different types of database were created, including a social economic database and a roadway characteristics database. Ten years social economic data, from 1995 to 2005 were collected for each of the 67 counties in the state of Florida, and a social economic database was created by manua lly imputing data obtained from different resources into the social economic database The roadway characteristics database was created by joining different GIS data la yers to the Tele Atlas base map. Based on literature review, in this study, the counties in Florida were divided into three groups based on the population in each county. The counties with the population less than 100,000 were classi fied into the rural area group. The counties with the population between 100,000 and 400,000 were clas sified into the small-medium urban area group. The counties with th e population greater than 400,000 were classified into the large metropolitan area group. In each group, two models were deve loped for estimating
58 the AADT values on Type II streets, one for state/county highway s and one for local streets. Stepwise regression method was used to se lect variables that will be included into the final models. All selected independent variables in the models are statistically significant with a 90% le vel of confidence. In total, si x linear regression models were built. The adjusted R2 values of the AADT prediction m odels vary from 0.166 to 0.418. Model validation results show that the MA PE values of the AADT prediction models vary from 31.99% to 159.49%. The author stud ied specific locations with large error. Some special urban facilities with more than two lanes were found to load traffic volume less than one thousand per day. ThatÂ’s why some large erro r caused. This problem may be caused by misclassification of road f unction or missing other potential important variables. The model with th e lowest MAPE value is found to be the minor state/county highway model for rural area. The model with the highest MA PE value is found to be the local street model for large metropolitan area. In general, minor state/county highway models provide more reasonable AADT estimates as compared to the local street model in terms of the lower MAPE values. 6.2 Final Result The major result of this st udy is the AADT values assigne d to the street segments in Florida counties where traffic counts are not available. The linear regression models developed in this study were used to estimate AADT values on Type II streets. So far, we have finished assigning AADT values to all Ty pe I and Type II streets for 67 counties in Florida, which account for about 93% of the streets in those counties. The estimated AADT values were merged to the Tele A tlas GIS base map based on the Dynamip_ID.
59 A .dbf file which contains all the informati on provided by the Tele Atlas base map plus the AADT values assigned to ea ch street segment was create d for each of the 67 counties. A picture for the final DBF file for Palm Beach County is given in Figure 6.1. Figure 6.1: The DBF File for Palm Beach County As mentioned before, traffic counts on Type III roads are extremely limited, and the samples available to this study are too small to build AADT prediction models with acceptable precision level. The linear regressi on models developed in this study provided tools for estimating AADT values on Type II st reets. However, some of the models suffer from large prediction errors in terms of the large MAPE values. It was found that minor state/county highway models provide more reasonable AADT estimates as compared to the local street model because local streets ha ve much fewer traffic counts available. A possible solution to these problems is to conduc t large-scale field da ta collection on Type
60 II and Type III roads to gather more AADT da ta. The collected AADT data can be used to develop AADT prediction models for Type III roads and re-c alibrate the local street model for Type II roads. The authors recomm end that future study could focus on these issues.
61REFERENCES 1. Q, Xia, F, Zhao, L.D Shen and D. Ospi na. Â“Estimation of Annual Average Daily Traffic for Non-State Ro ads in a Florida CountyÂ”. Research Report, Florida Department of Transportati on, Tallahassee, FL, June 1999. 2. F. Zhao and S. Chung. Â“Estimation of A nnual Average Daily Traffic in a Florida County Using GIS and RegressionÂ”. Tran sportation Research Record, No. 1660, Transportation Research Board, National Research Council, Wa shington, D.C, 1999, pp. 32-40. 3. D. Mohamad, K.C. Sinha, T. Kuczek and C.F. Scholer. Â“An Annual Average Daily Traffic Prediction Model for County Road sÂ”. Purdue University, Indiana, 1998. 4. N. Wang and H.L. Teng. Â“VMT Estima tion Associated with ITS Data and Maintenance of Loop Detectors.Â” University of Virginia, 2004. 5. S. Gadda, A. Magoon and K.M. Kockelman. Â“Estimates of AADT: Quantifying the Uncertainty.Â” Transportation Re search Record, January 2007. 6. J.K. EOM, M.S. Park, T.Y. Heo and L.F. Huntsinger. Â“Improving Prediction of Annual Average Daily Traffic for NonFreeway Facilities by Applying Spatial Statistical Method.Â” Journal of the Transportation Resear ch Board. No.1968. North Carolina State University, 2006. 7. Smith, B. L and Demetsky, M. J. Â“Tra ffic Flow Forecasting: Comparision of modeling approaches.Â” Journal of Transpor tation Engineering, American Society of Civil engineers, 261-6, 1997. 8. M. McCord, P. Goel, Z.J. Jiang, B.J. Coif man, Y.L. Yang, and C. Merry. Improving Â“AADT and VDT Estimation with High-Resolution Satellite Imagery.Â” American Society for Photogrammetry and Remote Sensin, 2002 9. Y.F. Tang, William H.K. Lam, M. ASCE an d Pan L. P. Ng. Â“Comparison of Four Modeling Techniques for Short-Term AADT Foresting in Hong KongÂ”. American Society of Civil Engineers (ASCE), 2003. 10. M.R. McCord, P.K. Goel, Z.J. Jiang, a nd C.Y. Park. Â“Improved AADT Estimation on Coverage Count Segments from Volume Correlations with Multiple ATR-Equipped Segments: Empirical Results from Ohio Highways.Â” Transportation Research Board Annual Meeting 2006 Paper. No 06-1296.
62 11. P.K. Goel, M.R. McCord and C.Y. Park Â“Exploiting Correlations between Link Flows to Improve AADT Estimation on C overage Count Segments: Methodology and Numerical Study.Â” 2005 Journal of the Tran sportation Research Board. No. 1917. 12. National Academies Keck Center. Washington, D.C. Traffic Monitoring Data, August 2007. 13. F. Zhao, M.T. Li and L.F. Chow. Â“Alte rnative for Estimating Seasonal Factors on Rural and Urban Roads in Florida.Â” Research Project. Florida International University, June 2004. 14. M. Aldrin. Â“Traffic Volume Estimati on from Short Period Traffic Counts.Â” Norwegian Computing Center. 15. United States Department of Transporta tion and Federal Highway Administration. Transportation Case Studies in GIS, September 1998.
63BIBLIOGRAPHY 1. Transportation Research Board. Highway Capacity Manual 2000. 2. American Association of State Highway and Transportation Officials. AASHTO Guidelines for Traffic Data Programs, 1992. 3. United States Department of Transporta tion and Federal Highway Administration. Traffic Monitoring Guide. Th ird Edition, February, 1995. 4. Florida Department of Transportation 2006 Annual Average Daily Traffic Report. 5. Palm Beach MPO. 2002 Traffic Count Program West Palm Beach Urban Study Area. 6. Palm Beach MPO. 2003 Traffic Count Program West Palm Beach Urban Study Area. 7. Palm Beach MPO. 2004 Traffic Count Program West Palm Beach Urban Study Area. 8. Florida Department of Transportation 2005 Lee County Traffic Count Report. 9. Indiana Department of Transportation. INDOT Traffic Monitoring System. 10. Virginia Department of Transportation and Federal Highway Administration. 2005 Virginia Department of Transportation Da ily Traffic Volume Estimates Including Vehicle Classification Estimates. 11. Transportation statistics Of fice. 1999 Florida Annual Aver age Daily Traffic Report.
65APPENDIX A. Descriptive Stat istics for Traffic Counts Table A.1: Descriptive Statistics fo r Traffic Counts in Hillsborough County Type Road Type FCC*Freque ncy Traffic Counts Percentage I Freeway A15 3595 309 8.60% Major US and State Highway A21 1971 217 11.00% A25 7699 560 7.30% II State and County Highways A30 155 9 5.80% A31 11067 891 8.10% A35 12573 832 6.60% A37 3 0 0.00% Local Streets A40 484 5 1.00% A41 75826 454 0.60% A43 2 0 0.00% A45 2922 85 2.90% III Vehicle Trail A50 12 0 0.00% A51 41 0 0.00% Ramp, Cul-de-sac, Traffic circle, Serve drive A60 2757 1 0.00% A61 1933 0 0.00% A62 217 0 0.00% A63 1306 8 0.60% A64 19 0 0.00% Driveway, Road in parking area, Alley A70 463 0 0.00% A71 138 0 0.00% A73 20 0 0.00% A74 1223 0 0.00% A75 508 0 0.00% Note: FCC: the functional classification code provided by the Tele Atlas digital map
66APPENDIX A. (Continued) Table A.2: Descriptive Statis tics for Different Types of Streets in Citrus County Type Road Type FCC* Freque ncy Traffic Counts Percentage I Major US and State Highway A21 902 131 14.5% A25 1252 180 14.3% II State and County Highways A30 3378 198 5.86% A31 163 8 4.91% A35 1251 36 2.88% Local Streets A40 8513 21 0.25% A41 17019 5 0.00% A45 172 0 0.00% III Vehicle Trail A50 3 0 0.00% A51 2 0 0.00% Ramp, Cul-de-sac, Traffic circle, Serve drive A60 133 0 0.00% A61 572 0 0.00% Driveway, Road in parking area, Alley A70 7 0 0.00% A71 1 0 0.00% A74 1223 0 0.00% Note: FCC: the functional classification code provided by the Tele Atlas base map
67APPENDIX A. (Continued) Table A.3: Descriptive Statis tics for Different Types of Streets in Nassau County Type Road Type FCC* FrequencyTraffic Counts Percentage I Freeway A15 105 4 3.81% Major US and State Highway A21 441 20 4.54% A25 1618 66 4.08% II State and County Highways A30 71 0 0.00% A31 2325 51 2.19% A35 1264 38 3.01% Local Streets A40 1523 0 0.00% A41 7057 0 0.00% A45 8 0 0.00% III Vehicle Trail A51 30 0 0.00% Ramp, Cul-de-sac, Traffic circle, Serve drive A60 144 0 0.00% A61 96 0 0.00% A63 26 0 0.00% A64 21 1 4.76% Driveway, Road in parking area, Alley A70 1135 0 0.00% A73 2 0 0.00% A74 122 0 0.00% A75 1 0 0.00% Note: FCC: the functional classification code provided by the Tele Atlas base map
68APPENDIX B. Type I Road Assignment Figure B.1: The First Step of Type I Roads AADT Assignment in Flagler County
69APPENDIX B. (Continued) Figure B.2: The Second Step of Type I Roads AADT Assignment in Flagler County
70APPENDIX B. (Continued) Figure B.3: AADT Assi gnment on Highway 100
71APPENDIX B. (Continued) Figure B.4: AADT Assignment on I-95
72APPENDIX B. (Continued) Figure B.5: AADT Assignment on State Highway 5
73APPENDIX C. Type III Roads The Type III streets only account for 5% in the whole network. They mainly include: Vehicle Trail, Ramp, Cul-de-sac, Tra ffic circle, Serve drive, Driveway, Road in parking area, and Alley. Traffi c counts on these types of roads are extremely limited, and the samples available in this study are too sm all to build an AADT prediction model. Due to this reason, we feel that it is very ha rd to estimate the AADT values on these streets without large-scale fiel d data collection. To demonstrat e the conclusion we made on the Type III streets, great efforts have been made to take a large number of field aero photos from GOOGLE EARTH. Based on large scale observation, it is f ound that Type III roads are composed of various assistant streets. It is improper to apply linear regression to predict AADT for these roads for the following reasons. 1) Too limited traffic count stations are availa ble on these roads. That means sample size for regression analysis is insufficient. 2) Type III roads vary dramatically on road characteristics and function, and traffic volumes carried on Type III roads are incomparable between each other. That means the error term is not identically distributed. 3) Traffic volume on Type III roads strongly re lies on utilities n earby, rather than roadway parameters or social economic factors. That means there is no strong linear relationship between AADT on Type II roads and the independent variables we have selected in this study.
74APPENDIX C. (Continued) Table C.1: Local Street FCC Description A40 Local, neighborhood, rural road city street, major category A41 Local, neighborhood, rural ro ad, city street, unseparated A42 Local, neighborhood, rural road, c ity street, unseparated, in tunnel A43 Local, neighborhood, rural road, c ity street, unsepar ated, underpassing A44 Local, neighborhood, rural road, city street, unseparated with rail line A45 Local, neighborhood, rural ro ad, city street, separated A46 Local, neighborhood, rural road, c ity street, separated, in tunnel A47 Local, neighborhood, rural road, c ity street, separated, underpassing A48 Local, neighborhood, rural road, city street, separated, with rail line Table C.2: Vehicular Trail FCC Description A50 Vehicular trail, road (4 WD) vehicle, major category A51 Vehicular trail, road (4WD) vehicle, unseparated A52 Vehicular trail, road (4WD) vehicle, unseparated, in tunnel A53 Vehicular trail, ro ad (4WD) vehicle, uns eparated, underpassing Table C.3: Ramp and Circle FCC Description A60 Access ramp, not associated with a limited-access highway A61 Cul-de-sac, the closed end of a ro ad that forms a loop or turn around A62 Traffic circle, the portion of a road that forms a roundabout A63 Access ramp, cloverleaf or limited-access interchange A64 Service drive, provides acce ss to businesses and rest areas
75APPENDIX C. (Continued) Table C.4: Other Facility FCC Description A70 Other thoroughfare, major category A71 Walkway, nearly level road for pedestrians, usually unnamed A72 Stairway, stepped road for pedestrians, usually unnamed A73 Alley, road for service vehicles located at the rear of buildings A74 Driveway A75 Road, parking area Figure C.1: A40 Unname d Street in Brad C ounty (Local Street)
76APPENDIX C. (Continued) Figure C.2: Bismark Ro ad (Local Street) Figure C.3: 4wd Road (Vehicular Trail)
77APPENDIX C. (Continued) Figure C.4: Trail (V ehicular Trail) Figure C.5: Trail 2 (Vehicular Trail)
78APPENDIX C. (Continued) Figure C.6: Ramp (Ramp) Figure C.7: Connecting Road (Ramp)
79APPENDIX C. (Continued) ` Figure C.8: Minnesota Road (Circle) Figure C.9: Suntr ee Road (Circle)
80APPENDIX C. (Continued) Figure C.10: Lake Andrew (Roundabout) Figure C.11: Diamond Ramp (Ramp)
81APPENDIX C. (Continued) Figure C.12: Service Road (Service Drive) Figure C.13: Driveway
82APPENDIX C. (Continued) Figure C.14: Park Area
83APPENDIX D. County Group Table D.1: County Gr oup based on Population ID Group County Name Population 1 Large Metropolitan Miami-Dade 2475388 2 Broward 1833871 3 Palm Beach 1290275 4 Hillsborough 1113288 5 Orange 1043057 6 Pinellas 972080 7 Duval 856085 8 Polk 531147 9 Lee 506395 10 Brevard 501814 11 Volusia 478425 12 Seminole 400380 13 Small-Medium Urban Pasco 384592 14 Sarasota 355972 15 Collier 321373 16 Escambia 315016 17 Manatee 298140 18 Marion 291154 19 Leon 264987 20 Lake 240896 21 Alachua 239804 22 Osceola 214215 23 St. Lucie 212907 24 Okaloosa 177289 25 Clay 157197 26 Charlotte 153873 27 Bay 153744 28 St. Johns 144096 29 Martin 142393 30 Hernando 141550 31 Santa Rosa 131376 32 Citrus 128837 33 Indian River 128750 34 Highlands 100225
84APPENDIX D. (Continued) Table D.1 (Continued) ID Group County Name Population 35 Rural Area Monroe 77328 36 Sumter 71902 37 Putnam 71365 38 Columbia 64650 39 Nassau 64559 40 Flagler 59021 41 Jackson 49619 42 Walton 47587 43 Gadsden 46796 44 Hendry 44306 45 Okeechobee 41598 46 Suwannee 39585 47 Desoto 39370 48 Levy 39162 49 Hardee 33924 50 Wakulla 28615 51 Bradford 27994 52 Baker 24365 53 Washington 23323 54 Taylor 21067 55 Madison 20235 56 Holmes 19608 57 Gilchrist 16542 58 Dixie 15495 59 Hamilton 14881 60 Union 14451 61 Calhoun 14176 62 Glades 13487 63 Gulf 13274 64 Jefferson 12854 65 Franklin 11813 66 Lafayette 8001 67 Liberty 7642
85APPENDIX E. The Social Economic Database for Florida Counties Table E.1: The Social Economic Da tabase for Florida Counties (2005) County Population Mileage Vehicle Municipality Labor Force Income Retail sales Alachua 18527 1480 193498 119964 123350 18527 2627159800 Baker 16673 837 24849 5399 10859 16673 150177200 Bay 21472 1395 171698 102720 83838 21472 2085346000 Bradford 15992 404 26984 7557 11581 15992 140709800 Brevard 25338 3162 503902 344638 256536 25338 6032675600 Broward 22165 5115 1333056 2216047 946775 22165 24431648800 Calhoun 13263 547 13527 3045 5200 13263 82358600 Charlotte 27497 2486 166855 17478 62267 27497 1657418600 Citrus 128837 2617 163775 10064 52328 23520 1401444200 Clay 157197 1120 167949 16561 83887 24969 1796073800 Collier 321373 1292 283778 48096 147722 34132 5138781400 Columbia 64650 1184 66556 10480 28510 14667 591838000 Dade 2475388 8798 1560708 1297192 1151712 17344 26876917400 De Soto 39370 471 32104 6840 13952 12436 271248600 Dixie 15495 457 18352 1939 5562 18397 44554800 Duval 856085 3791 677061 849435 433512 22153 11476711200 Escambia 315016 2371 252673 55281 136817 21254 3620050200 Flagler 59021 813 76656 59239 29448 32430 401300400 Franklin 11813 368 12935 3339 5055 18663 95141400 Gilchrist 16542 557 42465 15836 20162 15680 299095400 Glades 13487 224 19830 3399 7055 16030 43645400 Gadsden 46796 867 10118 1732 4286 15700 2340000 Gulf 13274 317 16687 4958 6444 14400 63694800 Hamilton 14881 612 12376 3280 4626 8591 53086800 Hardee 33924 609 24286 10361 11721 7878 171138000 Hendry 44306 443 39574 11847 17202 8397 353976000 Hernando 141550 1833 159449 6912 57643 22926 1213458000 Highlands 100225 1640 104669 20962 40280 18244 808460800 Hillsborough 1113288 5338 970179 383109 596028 25837 15696675600 Holmes 19608 840 19487 4079 8284 16804 53215800
86APPENDIX E. (Continued) Table E.1 (Continued) County Population Mileage Vehicle Municipality Labor Force Income Retail sales Lake 240896 2064 294639 119337 117393 24510 2337504000 Lee 506395 4619 569830 255654 272784 29995 7565003200 Leon 264987 1321 226763 166943 139602 24441 2949617600 Levy 39162 1193 52175 9208 16001 17272 290598200 Liberty 7642 464 7884 903 3405 21764 17384800 Madison 20235 771 19465 4029 7266 10780 84085400 Manatee 298140 1279 305062 74477 149758 24457 3041818400 Marion 291154 3543 329312 55218 127360 20662 3243623000 Martin 142393 542 143251 19156 64498 28576 2201889800 Monroe 77328 507 105021 55036 44651 29882 1345789400 Nassau 64559 984 76563 16123 31979 27434 477384000 Okaloosa 177289 1379 182484 77150 97865 24992 2909015800 Okeechobee 41598 426 47804 5634 16810 15042 375502200 Orange 1043057 4034 938702 356124 573640 22880 13574486000 Osceola 214215 1230 198032 83475 114591 19487 1992096800 Palm Beach 1290275 3622 979450 863186 617272 23658 19330602000 Pasco 384592 2012 406426 41504 177748 21834 3570871800 Pinellas 972080 4060 784051 749006 475340 24546 13151777600 Polk 531147 4020 558005 220014 262336 20468 4929128600 Putnam 71365 2012 84820 14332 30526 18050 558579600 Santa Rosa 131376 1772 142945 13074 64378 25389 952465800 Sarasota 355972 2418 360041 103623 178463 27674 4930959200 Seminole 400380 1574 403673 227760 226608 30209 6002229400 St. Johns 144096 925 162811 19410 81144 34490 1628020000 St. Lucie 212907 1618 223133 147983 110016 23316 2186050200 Sumter 71902 658 68446 8222 27297 21143 284887800 Suwannee 39585 1405 49610 7111 16389 15178 309790400 Taylor 21067 859 24955 6603 8612 17668 186670000 Union 14451 294 12339 2098 90187 15236 29972600 Volusia 478425 3044 471475 438886 239707 23674 5210287000
87APPENDIX E. (Continued) Table E.1 (Continued) County Population Mileage Vehicle Municipality Labor Force Income Retail sales Washington 23323 1246 24827 5024 12429 21297 130457800 Jefferson 12854 615 15374 2379 6631 20653 73796200 Lafayette 8001 468 7873 1055 2715 13071 42899600 Indian River 128750 997 130766 46924 58055 27238 1752925000 Jackson 49619 1528 48248 17797 21124 12339 410321000 Wakulla 28615 871 32470 684 13677 21971 95548000 Walton 47587 1291 56730 7022 29664 24303 330914800