U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590

Skip to content
Facebook iconYouTube iconTwitter iconFlickr iconLinkedInInstagram

Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations

This report is an archived publication and may contain dated technical, contact, and link information
Publication Number: FHWA-RD-03-050

Surrogate Safety Measures From Traffic Simulation Models

PDF Version (903 KB)

PDF files can be viewed with the Acrobat® Reader®

3. Traffic Simulation Model Overview

Microscopic simulation models hold some promise for collecting surrogate measures of safety for intersections. Microscopic models typically simulate traffic systems on a vehicle-by-vehicle basis by updating position, speed, acceleration, lane position, and other state variables on time steps, such as on a seconds basis, as the vehicles interact with traffic signals, signs, other vehicles, and roadway geometrics. Some simulations allow use of even smaller time steps for more accurate behavioral analysis and/or use an event-driven structure for more computational efficiency. Microscopic simulations generally also include detailed modeling of traffic signal operations. Accurate modeling of traffic signals will be a requirement for derivation of surrogate safety measures. However, all microscopic traffic simulation models were designed assuming that drivers behave in a "safe" manner, but according to their particular driver behavior characteristics (i.e., aggressiveness for gap acceptance and lane changing). This is true in the real world also, but because of misjudgment and mistakes, crashes do occur. Any derivation of surrogate measures must account for this basic fact that simulations do not (currently) include crash occurrence.

Without yet assuming a particular form of the SSAM (i.e., internal enhancements or external processing of model outputs), the pertinent characteristics of microscopic simulations for this project are:

  • General features such as user base, stability, usability, model bugs, etc.
  • Behavioral modeling of driver/vehicle interactions.
  • Ability to extract detailed data from the simulation (application programming interfaces (APIs), output files, open source).
  • Ability to calibrate and select parameters of models.
  • Cost to modify source or outputs to support surrogate measures.


Microscopic models that are well used in the transportation community, with easy-to-run analyses will be preferred for adaptation for surrogate safety measures analysis. Features such as post-processing analysis tools, graphical network editors, and extensible components are preferred.

Behavioral Modeling

For evaluation of surrogate safety measures, microscopic simulations must model the key driver behaviors that produce opportunities for crashes. Those behaviors are mainly:

  • Car following.
  • Gap acceptance.
  • Lane changing.

All microscopic traffic simulation models include these behaviors with varying levels of resolution and realism. However, models with especially detailed, realistic behavioral components will be more amenable for use in later phases of this surrogate safety measures project. Some evaluation of the comparative strengths and weaknesses of the behavioral components of available models is provided.

Data Extraction

Almost all of the proposed and existing surrogate measures of safety require detailed information about vehicle/vehicle interactions that is not typically available to the end-user from microscopic simulation models. Microscopic simulations with fewer barriers to data extraction, such as providing APIs or configurable output files, would be more amenable for use in the later phases of this project.

Calibration and Parameter Testing

The derivation of surrogate safety measures from simulation models is dependent upon the parameters used in the behavioral and performance sub-models. The ability to calibrate, modify, and manipulate these parameters is a key characteristic of microscopic models amenable for use in the later phases of this project.


Making modifications to existing model structure, architecture, and Graphical User Interfaces (GUI), and adding outputs, adding inputs, and other features of customized software can be expensive. Microscopic simulations that have a nominal modification cost (such as a cooperative vendor willing to make modifications for free) will be more amenable for use in later phases of this project. Leverage of past government expenditures should also be considered (e.g., CORSIM investment).

These characteristics are evaluated in more detail against commonly available microscopic traffic simulation models in the following sections. The models reviewed are CORSIM, Verkehr in Stadten – simulation (VISSIM), Simtraffic, Paramics, HUTSIM, Texas, Wide-Area Traffic Simulation (WATSIM), Integration, and AIMSUN. There are other microscopic traffic simulation models available in the community, which are used primarily for research (65). Only those that are commercially supported to some degree were evaluated.

Some elements of the tables contain value judgments for a specific model characteristic (high, medium, low, possible). These judgments are the opinion of the authors and do not reflect any official FHWA opinions or policies. Information that was not available is marked as "NI." Attributes that are not applicable to a particular simulation model are marked "N/A." Some "yes" indications are asterisked, indicating that additional detail is available in the discussion section for that row of the table. The evaluation is not intended to be exhaustive and was limited by the funding available under this contract. It includes only those elements of microscopic simulations that were anticipated to impact surrogate safety assessment and collection of surrogate measures. Best efforts were made to verify the accuracy of ratings with the simulation model developers and reviewing available documentation (66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80).


Table 2 compares the simulation models included in this review for the following general model characteristics:

Source Code Available

Availability of source code makes the model easier to change, modify, and understand the underlying models. Of those reviewed, only CORSIM and Texas have available source code. The other models are commercial products that are the principal business of the developer organizations.

Interaction With External Codes

Developers that have linked their simulations to other external software modules have more experience and understanding of what would be required for an external link to an SSAM. For example, CORSIM has been linked to Adaptive Control System (ACS) prototypes and Split, Cycle, and Offset Optimization Technique (SCOOT), VISSIM has been linked to Virtual NextPhase, and AIMSUN has been linked to Sydney Coordinated Adaptive Traffic System (SCATS).

Post-Processing Analysis Tools

Simulations with existing post-processors are more likely built to allow the SSAM to operate independently from the main code. Paramics and AIMSUN have post-processing tools supplied with their software. Many users of CORSIM and VISSIM have built post-processors for model output (and share them in the user community), but "integrated" tools are not available with the software itself. Texas can output data to formatted text files for import into spreadsheets (62).

Graphical Network Editor

Allows models to be built, manipulated, and visualized more easily. Preferred for general ease-of-use. Most of the models include some form of a graphical editor (some easier to use than others).

Graphical Network Editor Extensible

An "extensible" network editor may allow external codes to be configured, parameters set, etc. in conjunction with network creation and configuration. Paramics claims an "extensible" network editor, part of their suite of APIs.

Runs on a Personal Computer (PC)

Important for ease-of-use and distribution to the widest range of users. All of the models reviewed run on a PC, either native or through emulation.


Simulations with an object-oriented structure would probably be easier to modify, enhance, and augment for SSAM functionality (of course, that does not guarantee that the object model is appropriate or useful).

Actuated Signals Modeled

The SSAM is intended for analysis of North American intersections, which includes actuated traffic signals (e.g., evaluation of three-phase versus four-phase signals with respect to safety). Simulations that explicitly model actuated signals are preferred for SSAM application. CORSIM includes explicit modeling of National Electronics Manufacturers Association (NEMA) eight-phase controllers (down to simulation at the assembly-code level), but lacks detailed modeling of the transition (transition is not necessarily required for SSAM analysis). VISSIM models all controllers using their Signal state generator (VAP) modeling language or provides a software interface to field software, including NextPhase, Vehrkers Systeme - Plus (VS-PLUS), SCATS, and NH-VOS (Netherlands – no acronym available). However, an issue with connecting directly to field firmware for accurate signal controller modeling is running simulations only at real-time speeds (i.e., a 4-hour (h) simulation takes 4 h to execute). AIMSUN also allows NEMA controller modeling through an external software interface.

Table 2. General simulation characteristics comparison.

Click to view alternative text

ni      No information

n/a    Not applicable

*      Software can output formatted text files for spreadsheet analysis

**     With use of VAP language

***   With use of hardware-in-loop

Behavior Modeling

Table 3 compares the simulation models included in this review for many driving behavior and detailed model characteristics. It also should be noted that the behaviors are all interrelated and the emergent behavior from the combination of the elements is also important.

Parameterized Gap-Acceptance Model

One of the key modeling elements of any microscopic simulation. Tunable parameters are required to assess sensitivity of the SSAM to the gap-acceptance procedure. All models reviewed included a gap-acceptance model with configurable parameters.

Parameterized Lane-Changing Model

One of the key modeling elements of any microscopic simulation. Tunable parameters are required to assess the sensitivity of the SSAM to lane-changing procedure. All models reviewed included a lane-changing model with configurable parameters.

Parameterized Car-Following Model

One of the key modeling elements of any microscopic simulation. Tunable parameters are required to assess the sensitivity of the SSAM to the car-following procedure. All models reviewed included a car-following model with configurable parameters (81).

Parameterized Turning Speed

The speed at which turns are made should be tunable by the user or variable based on turning radius, number of lanes, etc. It is conceivable that the turning-speed model could influence calculation of surrogate measures. SIMTRAFFIC claims a parameterized turning-speed model and VISSIM and Texas allows turning speed to be dependent on vehicle type and turning radius.

Reaction to Yellow

Modeling of a driver’s reaction to yellow is important to measure dilemma-zone performance. It could be important for calculation of surrogate measures if the reaction model is variable by driver type, vehicle type, etc. Most models reviewed have reaction "by driver type." Paramics lists its modeling capabilities "by driver." This implies a continuous scale of parameters, rather than a set of fixed parameters (one for each type). VISSIM has reaction models with specific driver-type parameter settings for both signal sequences with and without flashing signals (for both European and North American signalization approaches).

Variable Driver Reaction Time

Reflects the model’s ability to represent the delay experienced between the driver’s identification of a potential collision and the application of control measures (braking, acceleration, or lane change) to avoid collision. In the real world, drivers’ reaction times vary by experience, age, etc. HUTSIM is planning integration of nanoscopic modeling of driver reactions and Paramics models driver awareness.

Intersection Box Movements

For assessment of surrogate safety measures, it is important for the simulation to model movement of the vehicles in the intersection with significant fidelity. For example, for left turns, Texas models intersection movements as combinations of appropriately sized arcs from the center of the beginning lane to the center of the receiving lane.

Variable Acceleration (and Deceleration) Rate

Simulations should include modeling of different vehicle capabilities by vehicle type. Unrealistic DRs (and maximum DR distributions) may underestimate the true statistics of surrogate measures. This is included in all models that were reviewed.

Sight-Distance Limits

Models that limit the "look-ahead" distance of drivers when making decisions (or model the look-ahead distance by driver or driver type) can more accurately model the awareness of drivers in surrogate measure statistics. In addition, sight-distance limits can reflect the modeling of roadway obstructions, such as curves, crests, trees, buildings, etc. This may also apply to modeling of in-vehicle sight restrictions, such as those that occur when following a large truck. Most of the models reviewed lack sophisticated sight-distance limitation modeling. VISSIM has some modeling of both the number of vehicles to look ahead and a distance ahead to consider before making maneuvers (as do other models as listed in the table), but no occlusion effects are modeled. CORSIM has a sight-distance limit for vehicles at the stop bar to look ahead for vehicles conflicting with their movement in the intersection.

Rolling Yield

Accurate modeling of yield signs and locations will be crucial for accurate collection of surrogate measures. It is hypothesized that the SSAM will be used for safety analyses of yield operations versus stop or signalized operations. A "rolling" yield indicates that the yield operation can occur with a slowed vehicle that does not come to a complete stop before re-entering the traffic stream.

Vehicles Interact With Pedestrians

Pedestrian safety is of extreme importance to traffic engineers. Simulations that model vehicle interactions with pedestrians may have the ability to assess the pedestrian safety effects of various alternatives (82). VISSIM and Paramics explicitly model pedestrian movements in crosswalks during pedestrian timings.

Friendly Merging

Refers to the phenomenon where certain driver types slow or stop to allow vehicles to merge (more) safely, which occurs in the real world, as opposed to only modeling slowing or stopping in a reactive sense. Friendly merging indicates that the following vehicle can create a gap for a merging vehicle. CORSIM and VISSIM include such behavior and AIMSUN includes such behavior for ramp junctions.

Modeling of Multilane Merging Behavior

In many locations, it is typical for vehicles entering the mainline flow to cross the path of an oncoming vehicle traveling in the same direction as the intended direction of travel of the entering vehicle and start accelerating in the adjacent lane. In this way, the oncoming vehicle can continue at its current speed without having to break for the turning vehicle (the maneuver is considered courteous behavior). Simulation models that allow for such behaviors to occur will more accurately represent the conflict behavior of locales that experience high volumes of such behaviors with wide multilane arterials. VISSIM can model such behavior with preferred entrance lanes for particular driver types, but it is not dependent on the lane that the oncoming vehicle is in.

Modeling of Right-of-Way in Intersection

A significant issue for modeling conflict events is that some turning behaviors must produce braking events by the traffic that has the right-of-way (i.e., making a left turn in front of oncoming traffic) to be considered unsafe events. If a simulation model does not represent this behavior, the surrogates cannot be reasonably measured. For example, AIMSUN calculates the TTC at the beginning of a left-turn maneuver to determine if a gap can be accepted with reasonable braking by the right-of-way vehicle. Therefore, some gap-acceptance maneuvers will, by definition, produce conflict events.

Modeling/Recording of Maneuver Failures

Acceptance of a gap is one event that can cause conflict events. On the other hand, the "rejection of gap" events may also have a surrogate safety implication. Models that can record the rejection or "failure" of the gap-acceptance process could produce another surrogate measure of the distribution and number of rejected gaps. Models that can export gap-acceptance event details could also easily export gap-failure event details. For example, Texas can export a table of conflict "check" (inherently a rejection if not followed by an acceptance event) and acceptance events.

Parking Maneuvers

On-street parking (parallel and double parking) creates conflict situations, lane changes, etc. in the real world and has a significant safety impact. Simulations that model on-street parking maneuvers are preferred. CORSIM models parking as "randomly occurring on-street incidents of variable duration," rather than explicitly modeling actual vehicles stopping to park and then restarting their trip later.  The mean duration of parking events must be less than 100 seconds, and there must be more than 14 events per hour.

Modeling of Turn Signaling

One significant aspect of rear-end conflict events is the use of turn signals by drivers. How turn signals (i.e., lack of signaling) affect the car-following and lane-changing logics is important to assessing the frequency and severity of rear-end conflict events. Turn signaling is notably a difficult modeling phenomenon. AIMSUN, for example, models the "emergency" of a vehicle changing lanes in advance of a turn to determine how aggressive the vehicle will be in cutting off right-of-way vehicles to make its turn, which could be considered a form of implicit modeling of turn signaling. VISSIM models turn signals for lane changes (i.e., turn signals are always used and some drivers will open gaps to allow merging), but does not model the presence or absence of a turn signal at a right or left turn at a junction. In addition, the presence of a turn signal on a vehicle in an adjacent lane affects driver behavior.


U-turns frequently cause conflict situations and some locations experience high enough volumes of U-turn traffic that their impact on safety should be addressed (e.g., including U-turns to businesses at the intersection corner or to access a freeway on-ramp). Simulations that include modeling of U-turns are preferred.

Origins and Destinations at the Intersection Corners

Many conflict situations are created by vehicles not turning at the intersection itself, but rather going to and coming from businesses at the intersection corners (e.g., convenience stores, gas stations, restaurants). Simulation models that can represent detailed business access situations will be preferred to those that cannot simulate such situations. For example, CORSIM would have difficulty modeling such situations because each driveway would have to be modeled as a separate node (intersection) and the minimum link length is 50 feet (ft) (15.15m) (some access driveways could be less than 50 ft (15.15m) from the traffic signal). VISSIM is the strongest model in this area, since each driveway would not have to be modeled as a node. The VISSIM structure is links and connectors with priority rules for right-of-way.

Table 3. Behavior modeling comparison.

Click to view alternative text

ni          No information

n/a        Not applicable

*          with use of "micronode" logic; also note that micronode logic found to be inadequate

**         specific look-ahead distance limit can be specified, but no obstructions

***       model can include preferred entrance lanes, but not dependent upon oncoming vehicles

****     Table of conflict "check" events can currently be exported with developer assistance

^          mean duration of parking events must be less than 100s; mean number of events must be greater than 14

^^        requires special link coding

Data Extraction

Table 4 compares the simulation models included in this review for the following characteristics of extracting data from the simulation:

Vehicle States Exportable to File

Indicates that the simulation is capable of exporting all of the vehicle state variables (speed, location, acceleration, lane, following identification number (ID), etc.) to a file that could be processed by an external SSAM. CORSIM includes an API that allows this. Paramics, VISSIM, and AIMSUN allow vehicle state variables to be exported as well.

Published Animation File Format

Indicates that the simulation’s animation file format is known or published. This would be important if the animation file output included enough information to allow an external SSAM to produce meaningful estimates. The CORSIM format is known and available; the Paramics format is known, and VISSIM publishes a .BTX format that contains vehicle-state variables that include information that could be used for external visualization.

API Available

Simulations with defined APIs are more amenable to interaction with an SSAM without extensive modifications to internal code. APIs are available for a number of codes, although there are no standards.

Output File(s) Configurable

Configurable output files allow the simulation to hide or display certain statistics (or calculate or not calculate). This capability could be leveraged into easily displaying certain SSAM aggregated statistics or not.

Gap-Acceptance Events Exportable

Simulations that can export data based on the occurrence of events are preferred (rather than just a time history of state variables for all vehicles that SSAM would have to post-process). Texas, for example, can export conflict check details for accepting and rejecting gaps.

Gap-Rejection Events Exportable

The flipside of exporting gap-acceptance events is exporting gap-rejection events. Texas, for example, can export conflict checks that fail for gap acceptance.

Lane-Change Events Exportable

Simulations that can export data based on the occurrence of events are preferred (rather than just a time history of state variables for all vehicles that SSAM would have to post-process). The details of a lane change can indicate whether or not a conflict event has occurred.

Vehicle-State Variables Include X,Y Position

For estimating surrogate measures of conflict events, the x,y position is needed for each vehicle over time (absolute x,y, not just relative x,y, to the end of the link, for example, if an arterial is simulated and not just a single intersection).

Currently Includes Conflict Statistics

Simulations that already compute conflict statistics or produce certain surrogates would certainly be preferred to those that do not. Texas appears to currently include calculations closest to those desired for surrogate measures calculations. VISSIM also reports various TTC-related outputs (with the research version of the software license) used by various car manufacturers (BMW, DaimlerChrysler, Volkswagen, Ford) to test the traffic impact of automatic cruise control algorithms. These outputs are computed for vehicles traveling in the same direction (i.e., lane changing on multilane links and within merging zones at exits and entrances). Less experience is available with TTC calculations within VISSIM for conflict maneuvers at intersections.

Table 4. Data extraction capabilities comparison.

Click to view alternative text

ni          No information

n/a        Not applicable

*          vehicle state variables exportable

**         relative to link position, not absolute

***       with micronode logic enabled; only provides total conflict counts by approach

****     for car-following only; used for adaptive cruise control calculations

^          time-to-collision only

^^        TTC summary statistics by distance from intersection per approach

Calibration and Parameters

Table 5 compares the simulation models included in this review for the following characteristics related to user-selectable parameters:

Variable Time Steps

Simulations with tunable time-step length are preferred to those with fixed-time steps for evaluation of sensitivity of surrogate measures to time-step size. In addition, simulations with variable time steps have more robust behavior models. Significantly, CORSIM does not allow tunable time steps.

Time Steps <1.0 second (s)

The precision of evaluating surrogate measures relies on frequent state-variable updates. The time scales of decision-making for surrogate measure evaluation are on the order of fractional seconds. The simulation must allow modeling of this fidelity.  Many of the simulations include tunable time-step resolution.

Gap-Acceptance Criteria Change by Delay

Many drivers in the real world change their behavior based on how long they have been waiting (i.e., they accept smaller gaps and apply larger accelerations the longer they have waited to make a particular opposed movement). Simulations that model this behavior are preferred. AIMSUN and Paramics claim to model such functionality for crossing flows; VISSIM and CORSIM model gap-acceptance behavior for lane changes that is modified based on the distance to the required movement (AIMSUN and Paramics also model a type of urgency for lane changes as the decision point comes nearer).

Vehicle Length

The safety of particular conflicting maneuvers is dependent on the size of the vehicles involved. All of the simulations reviewed include vehicle length.

Vehicle Length Considered by Gap Logic

Surrogate measures based on the proximity of two vehicles in space and time are affected significantly if the vehicles are modeled as points rather than rectangles. Some animation results indicate that some models do not adequately consider vehicle length for gap acceptance, or the animation routines are not accurate enough to indicate that vehicles would not have collided in the real world.

Variable Headways

Different driver types maintain different headways between the vehicle they are following based on their level of risk acceptance. This must be reflected in the simulation for accurate representation of surrogate measures. All models include this feature to varying degrees.

Variable Queue Discharge Headway

Related to variable headways, as the queue dissipates at a traffic signal, different driver types react at different rates that may have an affect on surrogate measures (primarily rear-end conflict measures). All models except Integration, which does not adequately model intersection dynamics, include this feature.

Table 5. Calibration and parameters comparison.

Click to view alternative text

ni          No information

n/a        Not applicable

*          gap acceptance for lane-changes modified by distance to required maneuver point

**         although animation results indicate otherwise


Table 6 compares the simulation models included in this review for the cost required to make changes to support surrogate safety modeling. Each cost is labeled "high," "medium," or "low." A low cost indicates that the effort required is estimated to be less than a person-month. A medium cost indicates that the effort required is estimated to be less than three person-months. A high cost indicates that the effort required is estimated to be more than three person-months.

Cost to Modify API

The anticipated cost on a relative scale (high, medium, low) to modify, change, or upgrade the API(s) of the simulation to allow SSAM operation.

Cost to Modify Output

The anticipated cost on a relative scale (high, medium, low) to modify, change, or upgrade the output files or formats of the simulation to allow SSAM operation.

Cost to Modify Input

The anticipated cost on a relative scale (high, medium, low) to modify, change, or upgrade the input files or formats of the simulation to allow SSAM operation (if required).

Table 6. Modification cost comparison.

Click to view alternative text

ni          No information

n/a        Not applicable

Previous | Table of Contents | Next

Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000
Turner-Fairbank Highway Research Center | 6300 Georgetown Pike | McLean, VA | 22101