The SANDAG transportation model provides a systematic analytical platform so that different alternatives and inputs can be evaluated in an iterative and controlled environment.
SANDAG Transportation Modeling
SANDAG deals with many complex mobility issues facing the San Diego region, including the development of a long-range Regional Transportation Plan (RTP). Transportation and land use models perform a very basic yet vital set of functions. Models are the principal tools used for alternatives analysis, and they provide planners and decision makers with information to help them equitably allocate scarce resources. The SANDAG transportation model provides a systematic analytical platform so that different alternatives and inputs can be evaluated in an iterative and controlled environment.
For the 2050 RTP, SANDAG uses an enhanced four-step transportation model. Four-step models have been the standard in transportation modeling since the late 1950s, and they are used by nearly every MPO in the United States for the development of transportation plans, corridor studies, Federal Transit Administration New Starts proposals, and air quality analyses. The estimates of regional transportation related emissions analyses meet the requirements established in the Transportation Conformity Rule, 40 CFR Sections 93.122(b) and 93.122(c). These requirements relate to the procedures to determine regional transportation-related emissions, including the use of network-based travel models, methods to estimate traffic speeds and delays, and the estimation of vehicle miles of travel.
The four major steps of the transportation model include:
- trip generation;
- trip distribution;
- mode choice; and
- network assignment (highway and transit).
After a first pass through the four steps, a feedback process is used to send congested travel conditions back into trip distribution and through to assignment. After several feedback iterations, a final pass is made through the mode choice and assignment steps to reflect congested travel conditions in mode decision-making. Travel model results are then combined with additional post-process input and output functions to form the complete modeling chain. Additionally, a truck model is run parallel to the four-step model, and truck trip tables are merged with passenger vehicle trip tables for highway assignment and air quality procedures.
A trip generation analysis is the first step in the transportation modeling process. Trip generation estimates the average weekday number of trip productions and attractions, or trip ends, in the region based on land use and demographic information from the regional growth forecast. Over a 24-hour period, roughly the same number of trips will originate in a zone as are destined there. However, residential zones will generate primarily trip productions while nonresidential zones will generate primarily trip attractions. The production/attraction distinction is important for the trip distribution model discussed in the next section.
Trips ends are calculated by all forms of transportation- including automobiles, trucks, taxicabs, motorcycles, public transit, bicycling, and walking- starting and ending in each transportation analysis zone (TAZ) and are categorized by ten trip types:
- Serve passenger
SANDAG develops a set of TAZs to aggregate areas of homogenous land uses and simplify the computer power needed to run the model. The TAZ system simplifies the region from nearly 1 million parcels to 4,670 zones (plus 12 external zones representing Orange, Riverside, and Imperial Counties and Baja California, Mexico) resulting in a matrix of 22 million zone combinations instead of 1 trillion combinations using parcels.
The trip-generation model works by applying trip rates to TAZ-level growth forecasts. The model calculates each of the trip ends separately as trip productions and attractions. Trip production rates are expressed as trips per household and vary by trip type and structure type. Trip attractions are expressed as trips per acre of nonresidential land use or trips per employee. Trip attraction rates vary by trip type and land use category. Trip per employee rates are used in Centre City San Diego where land use densities are much higher than regional averages. The 2050 Regional Growth Forecast was used to produce trip-generation forecasts for the years 2008, 2010, 2015, 2018, 2020, 2030, 2035, 2040, and 2050. Trip generation rates were established by utilizing data from traffic generator studies and the Institute of Transportation Engineers (ITE) trip generation report, as well as expanding rates from the 1995 and 2006 Travel Behavior Survey and the 2001 Caltrans Statewide Travel Survey.
The model reduces future year person-trips by a small amount to reflect the increased use of teleworking and e-commerce. Reduction factors of were applied to selected trip purposes and land uses. Telework reduction factors depend on the likelihood that the land use type would have employee categories that could feasibly telecommute. Reduction factors start in the year 2020.
The truck model follows a process similar to the one followed by the person model. The model computes truck vehicle trips for heavy-duty trucks, including light heavy-duty, medium heavy-duty, and heavy heavy-duty trucks. The truck classifications correspond to the California Air Resources Board truck classifications used in the air quality model EMFAC. Trip production and attraction rates are expressed as trips per employee, and the rates vary by employee industry category.
After trip generation, trip distribution allocates and balances trip productions and attractions through a gravity approach based on trip end density and location. Trip distribution considers the distance between a trip ends that is based on the assumed highway and public transportation networks that are input for any given future year. The model is designed to modify trip patterns in response to new land use developments and transportation facility changes. For example, the opening of a new shopping center would shift trips from other nearby shopping areas to the new development. Another example would be the introduction of mixed-use development. In this case the model would yield shorter trip lengths by recognizing the increased opportunity for interaction between residential and commercial areas in the development.
More specifically, trip movements between zones are determined using a form of the trip distribution model known as the doubly-constrained, gamma-function gravity model. Inputs to the trip distribution model include zone-level trip generation forecasts by trip type, zone-to-zone impedances, and gamma function parameters by trip type and 4D category. 4D index categories attempt to define locations by their density, diversity, distance, and urban design characteristics. A high 4D index value represents areas that would be considered smart growth and would result in shortened trip lengths. In this way, the model is designed to reflect changing trip patterns in response to the types of new development in land use scenarios. The model also modifies trip patterns as new roadways are added.
A truck trip distribution analysis is performed in a similar manner, but it is used to distribute vehicle-trips rather than person-trips by purpose, as in the person model. The truck model also uses different distribution parameters by vehicle type, which are not segmented by 4D category.
The model is calibrated to match observed trip length frequencies from the 1995 and 2006 Travel Behavior Survey. Zone-to-zone impedances are a composite measure of peak and off-peak travel times and costs by highway, transit, and non-motorized modes.
At this point in the modeling process, total person-trip movements between zones are split into different forms of transportation by highway, transit, and non-motorized modes (bicycling and walking). The mode choice step selects the most likely form of transportation for each trip, based on access, traveler’s income, trip purpose, parking costs, fuel price, transit fares, travel time, and other time and pricing parameters.
Highway modes include drive-alone non-toll, drive-alone toll, shared-ride non HOV/non-toll, shared-ride HOV/non-toll, and shared-ride HOV/toll. Each HOV mode is further identified as either a two-person HOV or three-plus person HOV. Transit modes are differentiated by five ride modes (Commuter Rail, Light Rail, Bus Rapid Transit or BRT, Express Bus, and Local Bus) and three access modes (walk, drive, and drop-off). The mode choice model is designed to link mode use to demographic assumptions, highway network conditions, transit system configuration, land use alternatives, parking costs, transit fares, and auto operating costs. Trips between zone pairs are allocated to modes based on the cost and time of traveling by a particular mode, compared with the cost and time of traveling by other modes.
Income level also is considered, because lower-income households tend to own fewer automobiles and therefore make more trips by transit and carpooling. People in higher income households tend to choose modes based on time and convenience rather than cost. The mode choice model is calibrated using 1995 and 2006 Travel Behavior Survey trip tables by mode and income, as well as 2001-2003 Regional Transit Survey transit trip characteristics. Regional-level Census 2000 work-trip mode shares also were used to fine tune mode-share estimates.
Highway and transit travel times reflect highway congestion effects from the final iteration of the feedback loop. The model produces peak and off-peak period trip tables for vehicles and transit riders. The a.m. peak period is from 6 to 9 a.m. and the p.m. peak period is from 3 to 6 p.m. The off-peak period covers the remaining 18 hours of the day.
Non-motorized trips (pedestrian and bicycle trips) reflect the effects of land use, trip purpose, and competing transportation modes. However, estimation procedures do not allow non-motorized facility issues to be addressed. For example, bicycle paths are not explicitly coded and thus do not affect non-motorized trip forecasts.
Highway and Transit Assignment
During network assignment, the model places each trip on the most efficient auto, transit, or non-motorized path based on the mode of transportation that was chosen earlier. Highway assignment produces traffic-volume estimates for all roadway segments in the system. These traffic volumes are an important input to emissions modeling. Similarly, transit trips are assigned to transit routes and segments.
SANDAG loads traffic using TransCAD’s “Multimodal Multiclass Assignment” function. Before loading the traffic onto the network, the truck modes are combined with the passenger vehicle modes. Multi-class assignment allows SANDAG to assign the vehicle modes (as defined in the highway network section) in one combined procedure.
The highway assignment model works by finding roads that provide the shortest travel impedance between each zone pair. Trips between zone pairs are then accumulated on road segments making up minimum paths. Highway impedances consider posted speed limits, signal delays, congestion delays, and costs. The model computes congestion delays for each segment based on the ratio of the traffic volume to roadway capacity. Motorists may choose different paths during peak hours, when congestion can be heavy, and off-peak hours, when roadways are typically free flowing. For this reason, traffic is assigned separately for a.m. peak, p.m. peak, and off-peak periods. Vehicle trip tables for each scenario reflect increased trip-making due to population growth and variations in travel patterns due to the alternative transportation facilities/networks proposed.
Model accuracy is assessed by comparing model estimated traffic volumes with actual traffic counts obtained through the SANDAG traffic monitoring program and the highway performance monitoring system estimates of Vehicle Miles of Travel (VMT).
After completing the highway assignments, additional processing is needed. Adjustments are made for calibration error volume, HOV/managed lane volume, bus volumes, hourly distribution factors, Level of Service, and travel time.
For transit assignment, TransCAD software assigns transit trip origins and destinations onto the transit network via Transit Access Points (TAPs). TAPs are selected transit stops, usually no more than ¼ mile apart, that are used to represent where transit trips load and unload the transit system. Separate transit assignments are produced for peak and off-peak periods, walk and auto access, and local bus and premium service. These individual assignments are summed to obtain total transit ridership forecasts.
External transit trips coming into San Diego from outside the region also need to be added to the internal transit trips estimated by the mode choice model. Currently, few transit trips enter from the north or east. However, more than 20,000 transit trips cross the Mexican border each day (i.e. walk to MTS trolley and busses near San Ysidro). To account for these trips, an external transit trip table for the base year is developed from on-board transit ridership surveys and factored to future years based on border crossing trends.
For accuracy, transit ridership forecasts from the transit assignment model are compared with transit counts from the SANDAG transit passenger counting program to determine whether transit modeling parameters need to be adjusted.
Some of these comparisons of model estimated boardings with actual boardings include:
- System-level boardings, which may reveal transfer rate problems and lead to changes to the transfer wait time factor in the mode choice model;
- Boardings by mode, which may reveal modal biases and lead to changes in mode choice modal constants;
- Boardings by frequency of service, which may show biases that lead to changes in the first wait factor in the mode choice model; and
- A Centre City screenline crossing, which may lead to changes in parking costs and boardings by stop location.
Model Iteration and Equilibrium
Once these four steps are completed for the millions of trips in the region on an average weekday, the SANDAG model iterates the trip distribution and mode choice step and runs through traffic assignment again based on levels of congestion measured in the previous iteration. The iterations continue until all trips are assigned the most efficient path for their mode. Each step is sensitive to an extensive set of inputs used to prepare a model scenario.
TransCAD 5.0 is the transportation planning computer package used by SANDAG to provide a framework for performing much of the computer processing involved with modeling, and it is used for the trip distribution and assignment steps. ArcInfo, a Geographic Information System (GIS), is used extensively in the modeling process as well to maintain, manipulate, and display transportation, land use, and demographic data. SANDAG has written numerous customized programs that provide a linkage between TransCAD and ArcInfo. Other custom programs perform some modeling functions, such as trip generation and mode choice.
A number of data files and surveys are used to calibrate the transportation models. There are four major inputs to the transportation models:
- Growth forecast inputs used to describe existing and planned land use patterns and demographic characteristics (described in Technical Appendix 2);
- Survey information;
- Highway networks used to describe existing roadway facilities and planned improvements to the roadway system; and
- Transit networks used to describe existing and planned public transit service.
The transportation models make use of survey data to establish relationships between input variables and model-estimated results. For example, trip generation rates are applied to dwelling units from the growth forecasting process to determine the number of trips generated from residential areas. Data collection is costly and time consuming, so surveys are conducted relatively infrequently. This normally does not create a problem since underlying model relationships are relatively stable over time. Surveys used include:
- 1995 and 2006 Travel Behavior Survey
- 2001 Caltrans Statewide Travel Survey
- 2001-2003 San Diego Regional Transit Survey
- External Trip Surveys (2006 Interregional Travel Behavior Survey)
- Traffic Generation Studies
- 1991 San Diego Visitor Survey
- 2000 Census Transportation Planning Package
- 2010 Gateway Forecast
- 2002 Freight Analysis Framework
- 2010 San Diego Parking Inventory Study
The regional highway networks in the 2050 RTP include all roads classified by local jurisdictions in their general plan circulation elements. These roads include freeways, expressways, and the Regional Arterial System (RAS). The RAS consists of all conventional state highways, prime arterials, and selected major streets. In addition, some local streets are included in the networks for connectivity between zones.
The route improvements and additions in the 2050 RTP are developed to provide adequate travel service that is compatible with adopted regional policies for land use and population growth. All regionally significant projects are included in the quantitative emissions analysis. These include all state highways, all proposed national highway system routes, all regionally significant arterials, and all “other principal arterials” functionally classified by the Federal Highway Administration.
The networks also account for programs intended to improve the operation of the highway system, including High Occupancy Vehicle (HOV) lanes, Managed Lanes, and ramp metering. Existing and proposed toll facilities also are modeled to reflect time, cost, and capacity effects of these facilities. The State Route (SR) 125 South, SR 11, SR 241, and additional lanes on Interstate 15 (I-15) north of SR 78 as well as additional lanes on I-5 north of Vandegrift Boulevard are modeled toll facilities included in the Revenue Constrained Plan for the San Diego region.
In addition, several managed/HOV lanes are included in the Revenue Constrained Plan. Facilities with proposed Managed Lanes include I-5, I-15, and I-805; and SR-52, SR-78, and SR-94. Managed Lanes are defined as reversible HOV routes and HOV routes with two or more lanes in the peak direction. Additionally, one-lane HOV facilities that operate as two-person carpool lanes in the earlier years of the plan transition to three-plus person managed lanes after 2035. It is assumed that the excess capacity not used by carpools and transit on these facilities would be managed, so that single-occupant vehicles could use these lanes under a pricing mechanism. Traffic flows would be managed so that the facility would operate at service level D or better.
SANDAG maintains a master highway network from which a specific-year network between the years 2008 (the 2050 Regional Growth Forecast base year) and 2050 can be built. Networks were built and verified for air quality conformity and SB 375 analyses of the 2050 RTP and EIR.
A list of the major highway and near-term regional arterial projects included in the analysis, along with information on phasing their implementation, is included in tables A.2 and A.4, located in Appendix A. Locally funded, regionally significant projects also have been included in the analysis. These projects are funded with TransNet funds, a 20-year, half-cent local sales tax for transportation that expired in 2008; TransNet Extension funds, a 40-year, half-cent local sales tax extension approved by voters in 2004 that expires in 2048; and other local revenue sources.
SANDAG also maintains transit network datasets for existing and proposed transit systems. Most transit routes run over the same streets, freeways, HOV lanes, and ramps used in the highway networks. As a result, the only additional facilities that are added to the transportation coverage for transit modeling purposes are:
- Trolley and commuter rail lines
- Streets used by buses that are not part of local general plan circulation elements
- Transit guideways
BRT service will have stations similar to commuter rail and light rail, and operating characteristics midway between rail and bus service. BRT service will be provided by advanced design buses operating on HOV lanes, some grade-separated transit ways, and surface streets with priority transit systems. Rapid Bus service would also utilize advanced design buses but they operate primarily along arterials. They would use queue jumpers and traffic signal priority measures to keep them operating at higher speeds.
Once TransCAD transit networks have been built, TransCAD finds minimum time paths between TAPs. The following four sets of paths are created for modes:
- A.M. Peak-period local bus
- A.M. Peak-period premium service
- Midday local bus
- Midday premium service
Bus speeds assumed in the transit networks are derived from modeled highway speeds and reflect the effects of congestion. Regional and express transit routes on surface streets are assumed to operate faster than automobiles in congestion due to priority transit treatments. Higher bus speeds may result for transit vehicles operating on highways with HOV lanes and HOV bypass lanes at ramp meters, compared with those routes that operate on highways where these facilities do not exist. In addition to transit travel times, transit fares are required as input to the mode choice model. TransCAD procedures replicate the San Diego region’s complicated fare policies which differ among:
- Buses, which collect a flat fare of between $1 and $4, depending on the type of service;
- Trolleys, which charge $2.50 for all trips;
- SPRINTER, which charges $2;
- Street-Cars, which are proposed to charge $2.25;
- Commuter rail (COASTER), which has a zone-based fare of between $4 and $5.50;
- Proposed regional BRT routes, which are assumed to charge $3 or $5 (for express BRT); and
- Proposed Rapid Bus routes, which are assumed to charge $2.25.
Fares are expressed in 1999 dollars (consistent with household incomes from the 2050 Regional Growth Forecast) and are assumed to remain constant in inflation adjusted dollars over the forecast period.
Near-term transit route changes are drawn from the Regional Short-Range Transit Plan, which was produced in cooperation with the region’s transit agencies. Longer-range improvements are proposed as a part of the RTP development and other transit corridor studies. In addition to federal and state funded projects, locally funded transit projects that are regionally significant have been included in the air quality conformity analysis of the 2050 RTP. These transit projects also are funded with TransNet funds or other local revenue sources. Once network coding is completed, the transportation models are run for the applicable scenarios. A list of major regional transit projects included in the draft air quality conformity analysis, as well as information on phasing their implementation, is included in table A.3, located in Appendix A.