Multidisciplinary Center for Earthquake Engineering Research logo google logo
navigation bar
bridgesmall.gif (4301 bytes)MCEER/NCEER Bulletin Articles: Research

Seismic Vulnerability Assessment of Highway Systems

by Stuart D. Werner, John B. Jernigan, Craig E. Taylor and Howard H. M. Hwang

This article presents research conducted on the development of a seismic vulnerability assessment procedure for highway systems funded under NCEER’s Highway Project. It represents work in progress and the data and models presented for the Memphis highway system are preliminary. They are for illustration purposes only and are subject to change. A more complete version of this paper will appear in the proceedings of the "National Seismic Conference on Bridges and Highways," sponsored by the Federal Highway Administration and Caltrans, to be held on December 10-13, 1995 in San Diego, California. Comments and questions should be directed to Stuart Werner, Dames and Moore, at (415) 896-5858.

Framework of Seismic Vulnerability Assessment Procedure

General Description
The general Seismic Vulnerability Assessment (SVA) procedure for a highway system is shown in figure 1. The procedure involves four main steps, which are: (1) initialization of the SVA; (2) development of system SVA results for each scenario earthquake and simulation specified under Step 1; (3) incrementation of the simulations and the scenario earthquakes and repeat of Step 2; and (4) aggregation of the SVA results for all earthquakes and simulations. Key to this process is a GIS database, which comprises several modules that contain the data and models for implementing the various steps of the system SVA.

95OctFig1.gif (24619 bytes)

This SVA procedure has several desirable features. First, it would be carried out within a GIS framework, which will enhance data management, analysis efficiency, and display of analysis results. Second, the GIS data base would be modular, in order to facilitate the incorporation of improved data, procedures, and models, as they are developed from future research and development efforts. Third, the procedure would be able to consider the effects of uncertainties in the earthquake characterization, hazard models, and vulnerability models, and would have the capability of developing aggregate SVA results that could be either deterministic or probabilistic, depending on user needs. This range of results would facilitate the usefulness of the SVA for seismic retrofit planning, prioritizing, and criteria development for an existing highway system.

With this as background, the remainder of this section outlines the basic features of the GIS data base for the SVA procedure, together with each of the above four steps of the procedure that are listed above.

GIS Data Base
The GIS data base would contain the data, models, and methodologies for:

The various modules that comprise the data base are summarized in the paragraphs that follow.

System Module. The system module would contain the basic data needed to define the system for the subsequent steps of the SVA. These data would describe: (a) system network configuration and linkages; (b) roadway widths (number of lanes each way); (c) traffic flows, capacities, and volumes for the roadways within the system; (d) component types and locations; (e) origin-destination zones; and (f) any special characteristics of the system, such as roadways designated as critical for national defense or for emergency response.

Hazards Module. The hazards module would contain the data and models needed to evaluate the seismic and geologic hazards throughout the system. The geologic data might include: (a) locations, earthquake activity rates, earthquake magnitude potentials, and tectonic displacement data for major faults and/or seismic zones in the region; (b) locations and topographic information for hills or valleys near the system that could be prone to landslide; and (c) data describing regional geology and local soil conditions throughout the system that would be needed to estimate ground shaking and potential ground movement due to liquefaction, landslide, etc. In addition, this module would include: (d) ground motion attenuation relationships; (e) models for estimating local soil effects on ground shaking; (f) landslide, liquefaction, and fault rupture models; and (g) models for characterizing uncertainties in ground shaking and ground movement.

Component Module. The component module would contain the data and models needed to estimate component vulnerabilities and their uncertainties. In this, component vulnerabilities would be characterized using loss models and functionality models. Loss models would estimate direct losses (repair and replacement costs) due to earthquake damage to the components. Functionality models, which are needed for post-earthquake traffic flow analysis, would represent the number of lanes open to carry traffic along each roadway link in the system at various times after an earthquake, together with any reduced speed limits due to earthquake damage. Input data for developing these models would also be contained in this module. These data would include: (a) structural attribute data needed to evaluate the seismic performance of each component under various levels of ground shaking and ground movement; and (b) damage repair strategies, costs, and traffic impacts, including the number of traffic lanes to be closed during repair, the durations of these lane closures, and the reduced speed limits for traffic in the repair areas.

Traffic Management Module. The traffic management module would accommodate information on post-earthquake traffic management procedures for alleviating traffic congestion after an earthquake. For example, experience following the Northridge earthquake showed that emergency traffic management procedures implemented by transportation planners and engineers from the City and County of Los Angeles and from Caltrans were very effective in reducing traffic congestion from earthquake damage.

Traffic Flow Module. The transportation flow module would contain the traffic model and system analysis procedure to be used for estimating earthquake effects on traffic flows throughout the system, for a given post-earthquake system state. This would consist of a system traffic forecasting methodology that would estimate effects of each scenario earthquake and simulation on such quantities as travel times, travel distances, and travel paths. These quantities could be estimated on an overall system basis (which would serve as a rough indicator of overall system performance) and also between selected origin-destination zones in the system.

Step 1: Initialization of Analysis
With the development of the GIS data base for the system to be analyzed, the actual SVA itself could be initiated. This would incorporate two parts. The first part would involve the use of earthquake source models (including randomization models) contained in the hazards module of the GIS data base to define a suite of scenario earthquakes that could affect the highway system to be analyzed. Uncertainties modeled would include geographic location of the earthquake source. Other uncertainties that could in principle be incorporated at this stage could address: (a) magnitude range for each source; (b) magnitude vs. fault rupture relationship; (c) orientation of rupture source; (d) directivity of rupture propagation; and (e) earthquake model uncertainties (e.g., uncertainties in "a" and "b" values in the Gutenberg-Richter relationship, or in characteristic earthquake models, time-dependent or time-independent models, etc.).

The second part of Step 1 would identify an adequate number of simulations for each scenario earthquake. In this, a "simulation" is defined as a complete set of hazards and component input parameters, in which the values of the parameters have been changed in accordance with the uncertainty characterizations contained in the hazards and component modules. For each simulation, a separate system SVA would be carried out under Step 2 (as described below). This process would be repeated until a sufficient number of simulations have been considered for each parameter to permit an evaluation under Step 3 of how the system SVA results are impacted by uncertainties in the parameter values. Effects of the uncertainties in each parameter can in principle be treated in this way. In what follows, each simulation for earthquake m (m=1,2,....M) is designated as n(m), (where n(m) = l(m), 2(m)....N(m), and N(m) is the total number of simulations for earthquake m).

Step 2: System Analysis for Earthquake m and Simulation n(m)
The next step in the SVA procedure would consist of a system analysis for each simulation associated with each scenario earthquake. For each simulation, the analysis would involve the following evaluations: Hazard Evaluation. First, the data and models contained in the hazards module of the GIS data base would be used to estimate the earthquake ground motions and geologic hazards throughout the system.

Direct Loss and System State Evaluation. Once the hazards are estimated, the component module in the GIS data base would be used to evaluate the direct losses and the system state associated with the mth earthquake and the n(m)th simulation for that earthquake. The direct losses would indicate the total cost for repair or replacement of damaged components within the system. The system state (defined at various times after the earthquake) would indicate the number of lanes that remain open to traffic along each roadway in the system, and any reduced speed limits within the system while the damage is being repaired.

Traffic Flow Evaluation. The system traffic flow models and traffic forecasting methodology contained in the traffic flow module would be applied for each system state, in order to assess how travel times, travel distances, and travel paths throughout the system and between its origin-destination zones would be impacted by the earthquake damage associated with the given system state. In principle, a local or regional socioeconomic model could be added at this stage, to evaluate broader social and economic impacts of the earthquake damage.

Step 3: lncrementation of Simulations and Scenario Earthquakes
This step simply represents the process wherein the system analysis from Step 2 is repeated for each simulation associated with each scenario earthquake.

Step 4: Aggregate System Analysis Results
This final step in the SVA process would be carried out after the system analyses for each simulation and each scenario earthquake have been completed. In this step, the results from all simulations would be aggregated and displayed. Depending on user needs, these aggregations could focus on the seismic risks associated with the total system or with individual components. Furthermore, the system or component results could be provided for individual simulations and/or for the broader (probabilistic) range of simulations. For research purposes, the impacts of incorporating variabilities into the SVA will be of considerable interest. For other purposes, such as the planning of seismic strengthening programs for existing highway systems, outputs can be adapted and/or simplified in accordance with the particular requirements of each user audience.

Demonstration Seismic Vulnerability Assessment
The above SVA procedure has been applied to a Memphis area highway system (fig. 2) in conjunction with currently available data and models, to demonstrate the application of the procedure and the type of results that can be obtained, and to also provide a basis for identifying and prioritizing research needs to be addressed in subsequent years of the NCEER Highway Project. It is noted that, because of the preliminary nature of much of the currently available data and models, the results of this demonstration SVA not be interpreted as a prediction of the seismic performance of this Memphis area highway system at this time. As improvements to these data and models are developed under the Highway Project, the reliability of the system seismic performance estimates should increase substantially.

95OctFig2.gif (33908 bytes)

The city of Memphis is located in the southwestern corner of Tennessee, just east of the Mississippi River and just north of the Tennessee-Mississippi border. Because of its proximity to the New Madrid seismic zone, the potential seismic risks to the Memphis area are well recognized and have been studied extensively (e.g., A & H, 1982; Desmond, 1994). The Memphis area highway system evaluated under this demonstration SVA (fig. 2a) includes the beltway of interstate highways that surrounds the city, the two crossings of the Mississippi River (at Interstate Highways 40 and 55), major roadways within the beltway, and highways just outside of the beltway that extend to important transportation, residential and commercial centers to the south, east, and north. Locations of such centers within this system are shown in the Origin-Destination (O-D) zone map provided in figure 2b. The system contains a total of 286 bridges.

As noted earlier, this demonstration SVA is based on currently available data and models only. Because the data and models are very preliminary at this time, it has been necessary to incorporate certain simplifying assumptions into this SVA. These assumptions are summarized below.

Scenario Earthquakes. This demonstration SVA was carried out for four scenario earthquake events only. These four earthquakes represent a range of different moment magnitude levels and locations in the region surrounding Memphis (fig. 3a), and are as follows: (a) Earthquake A-which has a moment magnitude MW = 7.5 (corresponding to a repeat of the largest earthquake in the 1811- 1812 sequence) and is located at the southern end of the New Madrid seismic zone (Zone A in fig. 3a); (b) Earthquake B -which has a moment magnitude MW = 6.5 and is located near the center of Zone A; (c) Earthquake C - which has a moment magnitude MW = 6.0 and is located in Zone B to the west of Zone A; and (d) Earthquake D - which has a moment magnitude MW = 5.5 event and is located in Zone B east of Zone A. The distances from the assumed epicenters of these various earth quakes to the closest and furthest points within the Memphis highway system range from about 35-50 km (for Earthquake D) to about 110-125 km (for Earthquake C). These earthquakes are described further in Werner and Taylor (1995). This article provides results for Earthquake D only.

95OctFig3.gif (44650 bytes)

System Module. The only system components that have been considered in this demonstration SVA are bridges and roadways. The system does not contain any tunnels, and other system components (e.g., retaining walls, etc.) have not been considered. The configuration and layout of the highways within the system were obtained as part of a GIS database provided by the University of Memphis. An extensive data base of structural attributes relevant to seismic performance have been compiled for most of the bridges in the area, and are provided in Werner and Taylor (1995). The traffic flow and volume data, roadway traffic capacities, and O-D zones within the system were provided by the Memphis and Shelby County Office of Planning and Development (OPD). The traffic flow data were from their 1988 traffic forecasting model.

Hazards Module. The only seismic and geologic hazard that has been considered in this SVA is ground shaking. Potential hazards from liquefaction, landslide, and associated ground movement have not been included because of a lack of suitable data for carrying out such evaluations over a spatially dispersed region and for a range of scenario earthquake events. The ground shaking hazard was represented in terms of peak ground acceleration (PGA). It was estimated in two steps. First, bedrock accelerations at each bridge site due to each scenario earthquake were estimated using the attenuation equation developed by Hwang and Huo (1994). Then, effects of local soil conditions at each bridge site were represented by multiplying the bedrock accelerations by local geology factors developed by Martin and Dobry (1994) for various site categories and bedrock acceleration levels. This was based on the local geology mapping of the area carried out at the University of Memphis (fig. 3b), and is contained in the GIS data base for this demonstration SVA (Hwang and Lin, 1993; Tarr and Hwang, 1993). Figure 4 shows the resulting bedrock and ground surface peak accelerations for Earthquake D.

95OctFig4.gif (30717 bytes)

Component Module - Loss Model. In this demonstration SVA, loss models previously developed under the ATC-25 project for conventional highway bridges were used to estimate direct losses for each bridge in the system due to each earthquake (ATC, 199 1). In these models, the direct losses depend only on whether the bridge has simple spans or is continuous/monolithic; i.e., other bridge structural attributes that could impact seismic performance have not been considered.

Component Module - Functionality Model. Because of a lack of available and suitably compiled data pertaining to post-earthquake traffic flows, repair procedures, and repair times for a given type and degree of bridge damage, only a very simple functionality model could be used for this demonstration SVA. Accordingly, the functionality model that was used represents the number of lanes open at discrete times after an earthquake, as a function of PGA and the original number of lanes along the bridge. Two different models were developed in accordance with the ATC-25 conventional highway bridge designations-one for bridges with simple spans and one for continuous/monolithic bridges. Reductions in traffic speeds were not considered at this time. In addition, to illustrate that system performance can vary with time after the earthquake, functionality models were developed for two discrete times. The first was intended to represent a time shortly after the earthquake, before any repairs have been made but after undamaged bridges had been reopened and lane closures to accommodate immediate post-earthquake repair had been established. This time has been arbitrarily assumed to be three days after the earthquake (recognizing that this may be optimistic). The second time was assumed to represent a more extended time after the earthquake, when some bridge repair has been made and at least some lanes of the damaged bridges have been reopened to traffic. This time was arbitrarily assumed to be six months. These approximate functionality models do not represent all of the possible causes of bridge and roadway closure after an earthquake, nor do they consider alternative bridge repair strategies that may be employed (together with their associated costs, durations, and impacts on traffic flow). The functionality models that were used are discussed in detail in Werner and Taylor (1995).

Traffic Management Module. In view of the absence of information addressing emergency traffic management procedures that would be implemented in the Memphis area after a major earthquake, the traffic management module of the GTS data base was not included in this demonstration SVA.

Traffic Flow Module. Our analysis of the impacts of each scenario earthquake on traffic flows within the Memphis area highway system was carried out using the MINUTP traffic forecasting software (Comsis, 1994). This software was selected because it is the procedure used at the Memphis-Shelby County Office of Planning and Development (OPD), and all traffic data for the region was available in the input format for this software. Although this procedure appeared to provide reasonable results for the various cases that were run (Werner and Taylor, 1995), it has the significant disadvantage of not being compatible with our GIS data base. Because this greatly increased the efforts required to develop suitable input data for the traffic flow analyses and to interpret the analysis results, the identification (or development, if necessary) of a suitable GIS-compatible traffic flow methodology is recommended as a high priority SVA research area.

Results: Direct Loss Estimates, Scenario Earthquake D
In accordance with the ATC-25 model used in this demonstration SVA, direct losses due to damage to the system’s bridges are represented as a damage ratio, DMG (%), which is defined as the ratio of the repair cost for each bridge to its total replacement cost.

The damage ratios for each of the 286 bridges in the Memphis area highway system due to each scenario earthquake are tabulated in Werner and Taylor (1995). To roughly compare the relative effects of each earthquake on the direct losses throughout the system, average damage ratios were computed (averaged over all of the 286 bridges) for each earthquake. This article provides results for Earthquake D only, for which this average damage ratio was 37.4%. From the results provided in Werner and Taylor (1995), this damage ratio turned out to be much larger than that computed for Earthquake C (whose effects on the system were relatively minor), and was slightly larger than the average damage ratio computed for Earthquake B. Only Earthquake A, which was by far the most severe of the four scenario earthquakes considered, resulted in damage ratios that were larger (by a substantial amount) than those due to Earthquake D.

95OctFig5.gif (15573 bytes)Results: Traffic Flows, Scenario Earthquake D
Overview of Seismic Vulnerability Assessment Procedure. This seismic vulnerability assessment estimated how earthquake damage to the Memphis area highway system due to each scenario earthquake impacted traffic flows in the area. The analysis consisted of two parts. First, the PGAs estimated for each scenario earthquake were applied to the functionality models, in order to estimate the state of the system at times of three days and six months after each earthquake (in terms of the number of available lanes along each roadway in the system). Then, the effects of any reductions in the available lanes (due to earthquake damage) on traffic flows throughout the system were estimated by using the MINUTP transportation forecasting software, together with a regional traffic capacity and flow data base developed at the Memphis and Shelby County OPD. From this, travel times and distances throughout the system after each earthquake were compared to pre-earthquake travel times and distances (in which all travel times and distances are average values for a 24 hour period). Two sets of comparisons were made. One corresponded to an overall travel time and distance for the entire system, which are computed as the sum of the travel times and distances respectively between all origin-destination (O-D) zones in the system (fig. 2b). This set of comparisons provides an approximate measure of the impacts of each earthquake on overall system performance. The second set of comparisons involved a breakdown of these total travel times and distances for particular key O-D zones highlighted in figure 2b. These latter comparisons indicate the spatial distribution of the earthquake impacts throughout the system, and also show how travel to and from these important O-D zones are impacted by earthquake damage to the highway system.

System State Results. Based on the PGA estimates obtained throughout the system due to each scenario earthquake, together with the preliminary functionality models, the system state after each earthquake was estimated. This system state is defined as the number of lanes open along each link in the system. The pre-earthquake system state and example system state results for times of three days and six months after Earthquake D are shown in figures 5 and 6. These figures show that, although Earthquake D is the smallest of the four scenario earthquakes (MW = 5.5), the gross models used for this SVA estimate that the proximity of this earthquake to the northern segment of the Memphis area highway system results in extensive roadway and lane closures in this segment, with lesser impacts on other sections of the system.

95OctFig6.gif (23235 bytes)

Overall System Travel Times. Table 1 shows that, as a result of the estimated bridge damage due to Earthquake D, overall system travel times three days after the earthquake are nearly 34 percent larger than the pre-earthquake values. Six months after the earthquake, the bridge repairs within that time have reduced the overall system travel time; however it is still nearly 20 percent larger than the pre-earthquake value.

Table 1: Effects of Earthquake D on Total System Travel Times and Distances


Pre-Earthquake Value

T = 3 days

T = 6 months


Percent Increase over Pre-EQ


Percent Increase over Pre-EQ

Total vehicle hours traveled over 24-hour period (incl. congestion)

3.73 x 105

4.99 x 105


4.46 x 105


Total travel distance (mi) over 24-hour period

15.5 x 106

15.6 x 106


15.6 x 106


Note: T = Time after earthquake at which system-wide impacts are estimated.

Overall System Travel Distances. Table 1 shows that over-all system travel distances are not sensitive to the estimated bridge damage due to Earthquake D, despite the fact that the total number of trips estimated over a 24-hour period by MINUTP (solely on the basis of demographics) was nearly the same for the pre-earthquake system and for each scenario earthquake. This lack of change of travel distances, despite significant increases in travel times, is no doubt due to the types of more direct but less-time efficient routes that would need to be taken after an earthquake. For example, if faster but less direct routes along interstate highways and beltways that would ordinarily be used are closed because of bridge damage, slower but more direct routes along city streets with no damaged bridges would instead need to be used.

O-D Zone Travel Times. Table 2 shows that three days after Earthquake D, the travel times between the zones listed in the table are estimated to be, on the average, nearly 16 percent larger than those for the pre-earthquake system. The travel time increases due to damage from this earthquake are estimated to be largest for northernmost of the highlighted zones, which are at Shelby Farms (Zones 249 and 252), Bartlett (Zone 264), and the Covington Pike (Zone 274). Six months after Earthquake D, table 2 shows that the travel times to and from these zones have been reduced substantially, and are now only 5.3 percent larger than the pre-earthquake values.

O-D Zone Travel Distances. As for the overall system travel distances, the travel distances to and from the highlighted O-D zones are not sensitive to damage from Earthquake D (Werner and Taylor, 1995).

Table 2: Effects of Earthquake D on Travel Time to or from Designated Origin-Destination Zones (Over 24 Hour Time Period)







Travel Time (hrs)

Percent Increase over Pre-EQ Time

Travel Time (hrs)

Percent Increase over Pre-EQ Time

Government Center (downtown Memphis)













Medical Center































University of Memphis







President's Island (Port)







Memphis Airport







Federal Express







Mall of Memphis







Hickory Hall

































Shelby Farms




















Covington Pike














The authors wish to acknowledge Ian Buckle and Ian Friedland of NCEER and Masanobu Shinozuka of the University of Southern California for their encouragement and helpful suggestions, Abdul Razak of the Memphis and Shelby County OPD for providing traffic data and for his help with the MINUTP traffic flow analyses, and Edward Wasserman and his staff at the Tennessee Department of Transportation in Nashville, Tennessee for providing valuable bridge data, drawings, and reports from their files. Finally, the authors are grateful to the following Dames & Moore personnel for their significant contributions to this research: C.B. Crouse (seismic hazard analysis), Ahmed Nisar (numerical analysis), and Jon Walton (GIS applications).

Allen & Hoshall, Inc. (A&H), (1982) Central United States Earthquake Preparedness Project (CUSEPP)-An Assessment of Damage and Casualties for Six Cities in the Central United States resulting from Earthquakes in the New Madrid Seismic Zone, A & H, Memphis, TN.

Applied Technology Council (ATC), (1991) ATC-25-Seismic Vulnerability and Impact of Disruption of Lifelines in the Conterminous United States, ATC, Redwood City, CA.

Comsis, (1994) MINUTP Technical Users Manual, Comsis Corporation, Silver Springs, MD, July.

Desmond, T.P., (1994) Seismic Risk Study for Selected Lifelines in Memphis and Shelby County, G&E Engineering Systems, Inc., Oakland, CA, March.

Hwang, H.H.M. and Huo, J-R., (1994) ‘Attenuation of Ground Motion in Hard Rock in the Central United States, " Paper submitted for publication to Earthquake Spectra.

Hwang, H.H.M. and Lin, H., (1993) GIS Database for Mapping Seismic Hazards and Seismic Risk in Memphis and Shelby County, Tennessee, Center for Earthquake Research and Information, Memphis State University, Memphis, TN, July.

Martin, G.R. and Dobry, R., (1994), "Earthquake Site Response and Seismic Code Provisions, " NCEER Bulletin, 8(4), pp l-6.

Tarr, A.C. and Hwang, H.H.M., (1993), "GIS-Generated Seismic Hazard Maps for Memphis and Shelby County, Tennessee", Earthquake Hazard Reduction in Central & Eastern United States: A Time for Examination and Action, Proceedings of 1993 National Earthquake Conference, Memphis, TN, pp 173-181.

Werner, S.D. and Taylor, C.E., (1995) Interim Year 2 Report for Task 106 E-7.3. I: Demonstration Seismic Risk Analysis of Highway/ Roadway System in Memphis, Tennessee, Prepared for National Center for Earthquake Engineering Research, Buffalo, NY by Dames & Moore, San Francisco, CA, September.


NCEER Bulletin, October 1995, Vol. 9, No. 4

top.gif (1113 bytes)

  Contact Us  |  Acknowledgements   |  Disclaimer  |  Copyright© 2007 by the Research Foundation of the State of New York. All rights reserved.