A Network Demand Model for Rural Bypass Planning. Paper Number

Transcription

1 A Network Demand Model for Rural Bypass Planning Paper Number Dr. Michael D. Anderson Assistant Professor of Civil Engineering Department of Civil and Environmental Engineering The University of Alabama in Huntsville Huntsville, AL (256) (256) Dr. Reginald R. Souleyrette Associate Professor of Civil Engineering Department of Civil and Construction Engineering Iowa State University Ames, IA (515) (515)

2 Abstract The subject of this paper is the application of desktop geographical information systems (GIS), widely available digital data and travel demand modeling packages to streamline the rural area bypass planning process. Bypass studies require quantitative estimates of travel demand on major roadway segments. Clearly, if travel demand models were available in rural areas, they would be useful. However, the scope of a bypass study, while possibly requiring a month of staff time or more, does not generally allow for the development of a travel model for bypass purposes alone. This paper presents a GIS-based travel modeling approach using a single internal trip purpose and an economic based method to approximate external trips, both of which are designed to minimize data collection efforts. The paper concludes, with a comparison of projected traffic using both the GIS-based bypass planning approach and traditional non-travel model approach, showing that the GIS-based methodology provides consistent, reproducible results with a reduced level of effort. The GIS-based modeling approach produced bypass traffic forecasts in less than one week versus three months for the original forecasts and as desktop GIS is used for maintenance, analysis and display of data are greatly enhanced. In addition, an analyst using the methodology is able to quickly analyze traffic conditions for many different alignments and access scenarios. Key Words: GIS, Bypass Analysis, Travel Modeling. Acknowledgements The authors would like to thank the Iowa Department of Transportation, the Federal Highway Administration s Eisenhower Fellowship Program, and the Center for Transportation Research and Education for making this research possible. In addition, the authors would like to extend thanks to Zachary Hans and Christopher Monsere for their input on this paper

3 1 Introduction With the completion of the interstate and most rural major arterials in Iowa, attention has turned toward maintenance and improvement of the highway system. Key to improving Iowa s intercity network is upgrading many through-town, two-lane facilities into four-lane divided highways. However, right-of-way constraints and landuse impacts make it difficult or undesirable to upgrade through-town two-lane facilities, resulting in the need to consider highway bypasses. Currently, as many as 40 bypass studies are under various stages of planning. With staff resources outstripped by increasing expectations of decision-makers and the public in general, planners and engineers are seeking ways to optimize planning efforts. The subject of this paper is the application of desktop geographical information systems (GIS), widely available digital data and travel demand modeling packages to streamline the bypass planning process. A brief history of urban network modeling efforts in Iowa is relevant to provide the institutional and technical setting for current planning capabilities. As in most states, development of travel models began in the 1960s and early 70s. As funding was available for origin-destination surveys modeling proliferated. Models were developed for cities as small as 25,000 through city/department of Transportation (DOT) partnerships. In the late 1970s, funding levels declined, resulting in maintenance of models only for Metropolitan planning Organizations (MPOs) with populations exceeding 50,000. Unfortunately, most highway bypass projects involve rural communities with populations under 50,000, where travel models are either outdated or non-existent, and thus, unavailable to support planning studies resulting in bypass analysis that has been performed without their benefit. Bypass studies require quantitative estimates of travel demand on major roadway segments. Clearly, if travel demand models were available in rural areas, they would be useful. However, the scope of a bypass study, while possibly requiring a month of staff time or more, does not generally allow for the development of a travel model for bypass purposes alone. Over the last 10 years or so, transportation planners and engineers, both in the public and private sector have been exploring and developing approaches to use GIS in conjunction with travel demand modeling software. Commercial products are available that either foster integration of third party demand and GIS models (e.g., Model Manager 2000) or allow for total model development and use in a specialized GIS package (e.g., TransCAD). The work reported in this paper relies on linkages developed on behalf of the Iowa DOT and Federal Highway Administration by the Center for Transportation Research and Education (CTRE) at Iowa State University, although other packages (or stand alone systems) could also be used. The paper examines alternative model development methodologies and recommends a GIS-based approach, minimizing data collection efforts. The model deploys a single internal trip purpose, and explores the use of an economic based method to approximate external trip distribution requiring minimal data collection. The paper concludes with a comparison of projected traffic using both the GIS-based bypass planning approach and traditional, non-travel model approach for a bypass around a small urban community in Iowa

4 2 Alternative Methodologies A report developed by Khisty and Rahi in 1987 identified 13 model development methodologies applicable for smaller urban areas as documented in transportation literature between 1963 and 1984 (1). Efforts peaked with the publication of NCHRP 187 in 1978, which reported equations and transferable parameters for communities with population exceeding 50,000 (2). Low (1972) suggests an approach to travel modeling that determines traffic volumes one link at a time, as a function of the relative probability that one link will be used in preference to another (3). In this model, interzonal trip probabilities are assigned to the network, producing estimated trip probabilities on a link-by-link basis. These individual link probabilities are used, along with actual traffic counts, to develop regression equations correlating interzonal trip probabilities and traffic counts. These regression equations are then used to estimate link volumes, using the interzonal trip probabilities obtained with horizon year trip generation results. In 1983, Neumann et al. proposed an alternate modeling approach directed towards smaller urban areas. The model distributes and assigns zonal socio-economic variables directly to the study area network (4). To use the model, external trips are removed, followed by the use of ground counts to develop a linear regression model as a function of socio-economic variables. As with Low s model, the regression model is again used to forecast horizon year traffic. Khisty and Rahi proposed yet another small area model referred to as an internal volume forecasting (IVF) model (5). This model followed the work of Low, while incorporating changes intended to reduce errors. One difference is the use of employees per zone and job positions per zone as the basis for trip generation. Another change is the use of total employment available in the study area as a factor in developing the trip interchange index assigned to the network. All three models (Low, Neumann et al., Khisty-Rahi) require high quality, system-wide traffic counts and assume stability of mathematical relationships between development and horizon years. For many rural communities, traffic count data are not readily available. Furthermore, trip generation equations and distribution parameters are likely to significantly change for small areas being bypassed. Therefore, a new model that is not dependent on existing traffic counts and is responsive to significant network changes is desired. 3 GIS-Based Model Development Methodology This section presents a GIS-based method for developing traffic models to study bypass projects. The approach forecasts traffic for the bypass and can display relevant turning movements for selected intersections. Multiple scenarios can be evaluated quickly and graphically (variations of alignment, access points, levels of access, and land uses). For this study, MapInfo was selected as the GIS platform due to ease of use, programmability and market penetration. Many, if not all of the functions described in this report can be accomplished using any number of GIS packages. However, additional programming would be required to achieve the same level of automation in other packages

5 With the tools developed by this research, users can quickly develop models, assign existing traffic demand to the network, check reasonableness of model outputs and make modifications if necessary. Subsequently, the model can be used to examine the implications of new bypass alternatives in base and horizon time frames. This section of the paper also documents the requirements for and availability of data for developing small area bypass models. It is followed by a demonstration of the development methodology in a case study for Waverly, a small Iowa community. The methodology is presented in a number of steps, which can be followed by an analyst who wishes to develop a similar model for any given study area. The section concludes by comparing results from the GIS-based approach to those resulting from conventional DOT methods for Carroll, Iowa. 3.1 Data Requirements and Availability To develop accurate travel models for rural areas, users are required to collect and maintain a limited number of data sets in GIS. These include roadway data from state or locally maintained cartographic databases, Census TIGER linework, BTS planning networks, or other sources. Average daily traffic linked to the graphic roadway elements is helpful for calibration and validation. The approach outlined in this paper requires a digital telephone directory (many available on CD ROM) with addresses for businesses and households in the community of interest. A travel modeling program is required to determine the volume of traffic on the roadways; Tranplan is used in this study. Street linework with street names and address ranges are required if the user needs to geocode any socio-economic data. Roadway data from the Iowa DOT s Coordinated Transportation Analysis and Management System (CTAMS) database include all public roadways in the state. For each segment, the database provides average daily traffic (useful in selecting roads to model and in validating the model), speed limits (used to determine initial travel speeds), and number of lanes (proxy for capacity). However, these data may be difficult or costly to obtain for many rural communities outside Iowa. In these cases, modelers must make certain assumptions on inclusion of roads in the model, speeds, and capacities. The source of socio-economic data required for model development is a digital telephone directory. These directories have addresses for most, if not all, businesses and households in the study area (referenced by city) which can be geocoded to the street network. Latitude/longitude coordinates are provided by some packages, allowing direct import to GIS, thus eliminating the need to geocode (address match) the data. If processing of the socio-economic data requires geocoding the information, a network with street names and address ranges is required. One source of attributed network data (networks with street names and address ranges) is the Census TIGER file series. A third party software package can be used to import TIGER files into MapInfo tables (see the MapInfo homepage for more information). Using these MapInfo/TIGER tables, the tabular, ASCII output file of socioeconomic data can be geocoded (information on how to geocode data can be found in the - 5 -

6 MapInfo Manual). The accuracy of the TIGER street network was tested for several Iowa cities and was determined to accurately geocode 55 to 70 percent of area households and businesses. If the analyst determines that additional geocoding effort is justified, manual improvements may be made in the address ranges in the network files (TIGER or other street map). Accuracy requirements should be determined on a case-by-case basis, however, for the rural areas studied, 55 to 80 percent accuracy was considered to provide a sufficient sample. The analyst will want to inspect the geocoded data to verify the avoidance of spatial bias in the sample (e.g., missing points are from only one section of town). Spatial bias must be addressed by manually adjusting or geocoding. 3.2 Model Development Steps: Case Study - Waverly, IA This section describes the steps for the model development methodology using Waverly, Iowa, with a population about 8,500. As with most small Iowa communities, the most recent origindestination study for the city was performed in the early 1970s. The steps described below would be similar for any bypass area, although community growth rate will affect the number of model scenarios needed. For example, if a community is not expected to experience significant growth in the horizon time period, only two models are required (with and without the bypass). If the community is experiencing slow growth or decline, a base year model should be developed and calibrated first, then two additional models should be developed (one for the horizon year with the bypass, and one for the horizon year without the bypass). For communities experiencing rapid or impending growth, up to four models in addition to a base year model may be required (two for the year of bypass opening with and without and two for the horizon year again, with and without the bypass). It is left to the discretion of the user to determine the number of models needed. The steps below can be repeated as many times as required, but are explained only in the context of developing a base year model without the bypass and a year of opening model (with the bypass). Both models assume the same trip generation as the base year (applicable in a community with stable travel demand characteristics) Step 1. Collect Data To begin model development, the analyst should collect the required data for the area of interest. All data should be imported into MapInfo. As mentioned, the CTAMS database covers roadway networks for all functional levels of Iowa highways. These roads are used to select the alignment of links to be included in the model network (again other data sets can be used during this step). Figure 1 shows the CTAMS cartography for the Waverly area. For non-iowa areas and for geocoding socio-economic data, a TIGER Street file or similar file is necessary. It is important to note that, especially for fast growing areas, TIGER or other street files may be out of date, and not include some streets that should be included in the model network

7 Socio-economic data are obtained through a Telephone Directory CD-ROM providing addresses for households and businesses in the study area. This data set is then addressed matched to the TIGER street file, if needed. For Waverly, the telephone directory identified 6753 locations for Using the 1994 TIGER data for Bremer county as the street address file, a match success rate of 55% was obtained using MapInfo s automatic geocoding procedure. Many of the remaining 45% percent of households and businesses could be manually placed using MapInfo s interactive geocoding procedures, although 55% provides an adequate sample without any obvious spatial bias. Collecting these data sets in MapInfo provides the basic data sets required for developing the model. It is important to note that these data sets can be collected and maintained in most any GIS package, but the automated model-creation programs developed as part of this study were written in MapBasic and operate only in MapInfo Step 2. Develop GIS Network After importing street and socioeconomic data into MapInfo, the analyst should establish traffic analysis zones and select the major streets in the network which will appear in the model. This step relies significantly on the planner s local knowledge and professional judgment. Zone location, number of zones, size and shape depend upon several rules of thumb as outlined in many transportation planning texts and in FHWA s Calibration and Adjustment of System Planning Models (6). To create a map of traffic analysis zones (TAZs), the analyst should create a new MapInfo table named TAZ.TAB (or other appropriate name). After developing the table, the user should make the table editable and use the polygon tool and define the traffic analysis zones through onscreen digitizing. Columns should be added for zone number, total households in the zone, and total businesses in the zone. It is important to number the zones sequentially from one to the highest numbered zone (Tranplan requirement). An example of traffic zones developed for Waverly is shown in Figure 2. The socio-economic data are then aggregated to calculate the number of households or businesses occupying a selected traffic analysis zone. Once address matching is completed, the aggregate totals are adjusted to represent 100% of households and businesses Step 3. Develop GIS Model in Tranplan Format After developing the TAZ table and compiling the socio-economic data, the next step is for the user to develop the model network in Tranplan format. To assist in this step, a MapBasic program, DEVELOP.MBX, was written to guide users through the process. Before running the DEVELOP.MBX program, it is important to copy it into the directory where the new model is going to be developed. It is also important that this directory not have any files named LINKS.* or NODES.*, as these filenames are reserved for use by the program (and will - 7 -

8 be overwritten if they already exist). To run DEVELOP.MBX, begin a MapInfo session and open any desired background tables (e.g., TAZ, TIGER linework, CTAMS streets, etc.). The user will want to carefully choose line styles and priority of the background layers to reduce clutter. After starting DEVELOP.MBX, a new MapInfo menu item will appear, MODEL_DEVELOPMENT, and the map window will indicate that two new tables, NODES and LINKS, have been created and opened. The user should then select the MODEL_DEVELOPMENT pull-down menu and choose the first option, LOCATE CENTROIDS. This option will provide the user instructions on the placement of the centroids, requiring that all centroids and external stations be located at one time and reminding the user that the centroids must be placed in an appropriate order. The user should next place points for all the centroids using the point (pushpin) tool and clicking the mouse button wherever a new centroid is to be located. Note that it is also important to locate the external stations at this point, but only after identifying all the internal centroids to ensure the program will correctly number the zones. After locating all the centroids and external stations, the user should select the next menu item, LOCATE NODES. Selecting this menu option will provide the user another message instructing them to locate the intersection nodes The user should then place nodes at the intersections of all network roads, and where centroid connectors connect to network streets, again using the point (pushpin) tool and clicking the mouse button where ever a node is to be located. The user MUST include all the nodes in the network during this step. It is recommended that user change the symbol style for node placement to facilitate quality control. Do not leave this step until you are sure that all nodes have been correctly placed. After placing the nodes, the user should select the LOCATE LINKS option from the menu. This will display a message to direct the user to heads-up digitize the roadway links. The user then should use the line placement tool and draw the links for the network. After drawing in the links, the network information is completed. Figure 3 shows a completed network. At this point, the user should select the final menu option, FINISH MODEL DEVELOPMENT. This will update the attribute information for the nodes and links and enter default values for speed and capacity. The default are speed of 15 mph with 0 ADT capacity for centroids and external stations and 30 mph and 6000 ADT one-way capacity for all other links. The user can change any of the default values, with area specific data. Finally, the menu is removed and the program will display a message informing the user to enter socio-economic data and run the model Step 4. Trip Generation After developing the model, the user must develop productions and attractions for the zones. The total number of households per zone from the TAZ table will be used to calculate the zone - 8 -

9 productions using a measure of trips per household. A common factor used is 9.2 trips per household, from NCHRP ITE s Trip Generation manual can also be used. The internal zone attractions are calculated as the percent of zone businesses to the total businesses in the entire study area balanced to the total number of productions in the study area. Caution must be applied here, as a single, large employer, or even a concentration of larger employers may tend to significantly skew the data. The user is cautioned to use as much local information as possible. For the case study, socio-economic data taken from a telephone directory CD-ROM were used to calculate productions using a factor of 9.2 productions per household. Attractions were calculated as the total productions times the number of businesses in a given zone divided by the total number of businesses. These calculations were performed directly by the software entered in the node table. The calculated values for Waverly are shown in Table Step 5. External Analysis At this point external-external (E-E) and external-internal (E-I) trip information is required. Huff s Probability Model is presented to develop splits between E-E and E-I for external zone traffic. Obviously, similar methods for developing the splits in traffic can be used. Huff's Probability Model is a probabilistic gravity model is formulated as a ratio of one city s attractiveness versus the summation of the attractiveness of all the other surrounding cities. The model is stated as: This model is useful in identifying the percent of vehicles from surrounding cities that would be attracted to the study area versus those that would pass through the study area to a competing attraction. Clearly, these percentages result in the amount of external-external traffic for the study area. The model can be applied directly or with the use of a spreadsheet to organize the data. An example of the model information is shown in Figure 4. The equation calculates the likelihood that a person living in one city shops in another, as well as the ability of one city to attract shoppers from surrounding cities. The information in columns (from Figure 4) represents, for each city, the likelihood that residents from that city shop - 9 -

10 in another city. The information in columns (from Figure 4) represents the ability of a city to attract shoppers from surrounding communities. Manual modifications should be made for trips not likely to travel through the study area. For example, if a trip between two cities can be made without passing through the study area, the entries for these trip values should be set to zero. Effectively removing external trips not expected in the study area. The likelihood of external-internal trips, calculated for each city, is a summation of those trips that are produced by the study city shopping in another town and the likelihood that another town s resident is attracted to the study city. If we use Oelwein for an example, the externalinternal trips for this city is the sum of cells E19, B22, E26, B29. For each city in the surrounding area, the likelihood of external-external trips, is the summation of all trips coming from and going to all other communities, through the study area. Again, continuing to use Oelwein for an example, the external-external trips for this city is the sum of cells E20, E21, E23, E27, E28, E30. With these values, the percent of external-internal and external-external trips can be determined. To convert the percentage of trips into the number of trips that are external-internal and external-external, the total traffic daily volume at the external station is multiplied by the appropriate percentage (it is important to select a traffic count value collected relatively close to town). This calculation will generate the total number of trips; however, for Tranplan, this information must be converted into productions and attractions. This conversion is performed by simply using 50 percent of the external-internal trips for productions and 50 percent for the attractions. The same 50/50 split is used for the external-external trips to generate external-external productions and attractions. As with the internal zones, the calculated productions and attractions will need to be entered into the node table. This table should already contain the production and attractions values for the internal zones. The external-internal production and attraction values for the external stations and external-external production and attraction values should be manually entered into the node table. This is shown in Table 2. For more information, please see the Huff documentation (7) Step 6. Execute Tranplan Model At this point, network information is stored in MapInfo as node and link tables and needs to have trip distribution and traffic assignment performed to forecast the traffic. A procedure to accomplish this has already been developed and programs and documentation are available from the University of Alabama in Huntsville or Iowa State University via the internet at either coeweb.eb.uah.edu/anderson/transgis/fhwa/index.htm or The output from Tranplan will be the assigned network traffic volumes. Figure 5 shows the traffic volumes obtained from running the model through Tranplan

11 It is recommended that the planner compare the results of the base model volumes to actual traffic counts from any available traffic count source. For the case study, the R-square difference between the model volume and the AADT was.90 and Figure 6 indicates the fit of the links. If this calculation results in a poor R-squared value, it is recommended that the analyst attempt to determine the reason for the poor validation, and make changes to improve the assignment. Changes to the model can include altering the roadway travel speeds, increasing the speeds for under-assignment and reducing the speeds for over-assignment. The changes in speed, however, should be justified by travel patterns in the community and the authors recommend that speed changes be made in five mile per hour increments and remain reasonable Step 7. Repeat Model Development Process with Bypass After the base model has been validated to the ground counts, the next step is to develop the bypass for the area and perform another run of the Tranplan software to determine the number of vehicles that would use the new bypass. It is recommended that the user create a new directory and copy all the MapInfo and Tranplan files into it. In MapInfo, first make the nodes layer editable and place new nodes at all the desired locations to complete the bypass. The user should then make the links table editable and digitize new links to the network to complete the bypass. Next, the user should run the MapBasic program BYPASS.MBX to update the LINKS and NODES tables with the new data for the bypass. At this point the user can either edit the default links attributes or perform a model run directly Step 8. Forecast Traffic for Horizon Year (or any year of interest) The final step is to develop forecasts for future traffic in the area. To perform this step, the planner should make estimates of future trip generation variables. Again, awareness of local growth issues is important. For areas expected the experience significant growth or decline, the user should perform this operation for both the base scenario and the bypass scenario to develop the final volumes for the area. This should also be done for any other access scenarios or alignments of interest. 4 Application of the modeling methodology The MapInfo-based bypass development methodology was applied for the community of Carroll, Iowa, as this area was recently studied using manual techniques, without the aid of a travel demand model. Projected traffic volumes resulting from the GIS-based methodology compared favorably to the results of the forecasting method used by the Iowa DOT. The differences, which generally range from less-than-ten to twenty percent are outlined in Table 3. Differences of this magnitude would generally not have serious implications for roadway design

12 5 Conclusions In conclusion, the proposed methodology for developing highway bypass plans within MapInfo provides consistent, reproducible results with greatly reduced level of effort (as compared with manual, non-travel demand model techniques). Using the GIS-based modeling approach, bypass traffic forecasts were obtained in less than one week, the original forecasts took three months. As the new model and data are maintained in a GIS, analysis and display of data are greatly enhanced. In addition, an analyst using the methodology is able to quickly analyze traffic conditions for many different alignments and access scenarios. Currently, right-of-way and landuse restrictions for roadway expansion conflict with travelers desire to decrease travel times causing many state DOTs to focus on highway bypasses as alternatives to two-lane through-town roadways. The GIS-based bypass model presented in this paper provides a methodology to support bypass planning decisions that relies on accepted travel demand modeling techniques developing impartial forecasts of future traffic to base infrastructure investment decisions. 6 Reference 1. C. J. Khisty and M. Y. Rahi. Inexpensive Travel Demand Techniques. Report WA-RD Washington State Department of Transportation, Olympia, WA NCHRP. Quick-Response Urban Travel Estimation Techniques and Transferable Parameters, Report 187. Transportation Research Board D. A. Low. A New Approach to Transportation System Modeling. Traffic Quarterly, Vol. 26, No E. Neumann, J. Halkias, and M. Elrazaz. Estimating Trip Rates for Small- and Medium- Sized Cites. Journal of Transportation Engineering. Vol. 109, No C. J. Khisty and M. Y. Rahi. Evaluation of Three Inexpensive Travel Demand Models for Small Urban Areas. Transportation Research Record, No D. Ismart. Calibration and Adjustment of System Planning Models. USDOT FHWA D. L. Huff. A Probabilistic Analysis of Shopping Center Trade Areas. Land Economics

13 Figure 1. CTAMS GIS data for Waverly

14 Figure 2. Waverly traffic analysis zones

15 Figure 3. Complete Waverly model

16 Figure 4. Example spreadsheet for Huff s Probability Model

17 Figure 5. Model volumes for Waverly

18 Figure 6. Validation of the Waverly model

19 Table 1. Internal-internal productions and attractions

20 Table 2. All production and attraction data for Waverly

21 Table 3. Differences between DOT and GIS methodologies for Carroll. Location DOT volume GIS-based volume Difference On northern bypass 1 east bypass 2,900 2, west bypass 2,100 2, On residual roadway 3 east of Carroll to 4,800 5, Monterry Drive 4 Monterry Drive to 6,950 8,197 1,247 Grant Road 5 Grant Road to 10,800 9,021-1,779 Main Street 6 Main Street to US 11,100 12,682 1, US 71 to Burgess 5,700 5, Avenue 8 Burgess Avenue to 3,500 3, west of Carroll