MCEER/NCEER Bulletin Articles: Research
Evaluation of Bridge Damage Data from Recent Earthquakes
by Anne A. Kiremidjian and Nesrin Basöz
This article presents research resulting from NCEERís Highway Project, task 106-E-7.3.3. It is based on a technical paper submitted for inclusion in an annual report to the Federal Highway Administration summarizing Research Year 3 of the Seismic Vulnerability of Existing Highway Construction project. Comments and questions should be directed to Anne Kiremidjian, Stanford University, at (415) 723-4164; email: email@example.com.
Data on bridge damage from earthquakes is becoming increasingly more available. Such data, however, have not been systematically studied to evaluate damage characteristics and correlate these to observed or estimated local ground motions. In this task, data on bridge damage from the Loma Prieta and Northridge earthquakes were studied to correlate observed bridge damage to:
- Structural characteristics of a bridge
- Local ground motions
- Repair cost
In order to achieve these objectives, statistics on structural characteristics of bridges, ground shaking levels at bridge sites, damage characteristics and repair cost were obtained. Next, empirical damage probability matrices and fragility curves were developed from data on bridge damage. In addition, correlation between structural characteristics and observed damage were determined.
Despite the high ground motion levels observed in the 1994 Northridge, California earthquake, only about 3% of all the bridges in the area experienced major damage. The analyses of bridge damage data showed that concrete structures designed/built with older design standards were more prone to damage under seismic loading. The ground shaking level, skew angle, abutment type, pier type and span continuity showed the highest correlation with observed damage. The total repair cost for the damaged bridges was about $150,000,000, with repair and/or reconstruction of the collapsed structures forming a large portion of the total.
Bridge Damage Data Analysis Method
Characteristics of the Database
The database compiled for the Loma Prieta and Northridge earthquakes consists of four main types of data: (a) structural characteristics, (b) bridge damage, (c) repair cost, and (d) ground motion levels and soil characteristics at bridge sites. A relational database management system (RDBMS), dBaseTM , was used to compile and perform queries for data on bridge damage and structural characteristics of bridges. In addition, a geographic information system (GIS), Arc/InfoTM , was used to obtain the ground motion levels at each bridge site.
Structural Characteristics: Structural characteristics compiled for the groups of bridges that were exposed to ground shaking included abutment type, number of spans, type of superstructure and substructure, length and width of the bridge, skew, number of hinges at joints and bents, abutment and column foundation types, retrofit history, and design year to represent design standards, such as column reinforcement and seat width. These structural characteristics were obtained from the Structural Maintenance System (SMS) database compiled and managed by Caltrans (Caltrans, 1993). In addition, more detailed information was obtained from Caltrans for some of the damaged bridges. Caltrans is currently in the process of compiling a database that includes information on abutment, bent/pier/column and bent/pier/footing details, such as seat width and type of bearings, footing type and column/footing connection. However, only about 15% of all the California bridges are currently in this database, thus this detailed information could not be used in the statistical analyses.
Bridge Damage: Detailed damage descriptions and the corresponding damage states were compiled for bridges damaged in the two earthquakes. The damage descriptions were obtained mainly from bridge damage reports compiled by Caltrans (1989, 1994). For the Northridge earthquake, these descriptions were cross-referenced with those provided by Buckle (1994), EERI (1995), and Yashinsky (1995). Judgment was used to treat inconsistencies in the interpretation of the observed damage data.
The database on bridge damage specifies two damage states (minor, major) for bridges damaged in the Loma Prieta earthquake, and four damage states (minor, moderate, major and collapse) for those damaged in the Northridge earthquake. The bridge damage data were used in correlation studies to obtain ground motion-damage relationships.
Damage State Definitions: Currently, no guidelines for evaluating physical bridge damage exist. The terms minor, moderate and major damage were subjective. Definitions of damage states for columns, abutments, and joints and connections for concrete bridges are proposed which were developed based on the observed bridge damage in the Northridge earthquake (Basöz, 1996). A questionnaire was prepared to acquire expert opinion on the proposed damage state definitions and given to bridge engineers at Caltrans. The feedback provided by the bridge engineers will be used to modify some of the damage state definitions.
Repair Cost: The estimated repair cost values for damaged bridges were obtained from supplementary bridge reports compiled by Caltrans following each earthquake. The database includes total estimated repair cost and more detailed information on repair work and cost for each bridge that was repaired. The repair cost ratio, defined as the ratio of repair cost to replacement cost of a bridge, was calculated for all the damaged bridges. The replacement cost of a bridge was estimated to be $90/ft2 based on the 1995 cost books.
Ground Motion Levels: In addition to structural characteristics, soil type at each bridge site and peak ground acceleration (PGA) levels observed in the two earthquakes were compiled. In order to obtain empirical ground motion-damage relationships for the set of bridges damaged in the Northridge earthquake, two sets of peak ground acceleration (PGA) values were used as the ground motion levels: (a) PGA values reported by USGS (1994), which were obtained from the contours of observed PGA recordings in the horizontal direction and (b) PGA values reported by WCFS (1995) that were obtained from the contours of the average of the PGA values measured in the E-W and N-S directions. Recorded ground motion levels were used to scale the parameters of empirical Greenís functions which were used in simulating the ground shaking levels (Somerville et al., 1996). The PGA value at a given bridge site was obtained within GIS by overlaying the ground shaking map and the bridge location map. Subsequently, the highest PGA values obtained at a bridge site were 1.55g and 0.66g for the USGS and WCFS maps, respectively. Since the PGA levels from the two data sets varied considerably, correlation studies were performed for both data sets. Because no contour maps were available for the PGA levels observed during the Loma Prieta earthquake, attenuation relationships were used to estimate the level of ground shaking at each bridge site.
Classification of Bridges
The compiled inventory of bridges was reviewed to: (a) select bridges to be used in correlation studies, (b) select structural characteristics (attributes) that best describe the seismic response of bridges, and (c) verify the correctness of the attribute values included in the bridge inventory database.
Data Sets: Several data sets were used for statistical analyses. All the analyses were performed for the state bridges since most of the reported damage in both earthquakes pertain to state bridges. First, all highway state bridges were selected and gathered in the highway bridge data set, and statistics on design year and ground shaking levels were obtained. Most of the bridge damage pertained to concrete structures. The number of damaged steel bridges was not large enough for statistical analysis. Therefore, concrete bridges were selected from the highway bridge data set.
One of the objectives of this research was to identify the effect of various structural characteristics on bridge damage. In order to study the effect of structural component types on bridge damage, such as effect of abutment type (monolithic or non-monolithic), and effect of number of columns per bent, bridges with single abutment type and one column bent type were compiled in a database called the homogeneous data set. That is, in order to determine the effect of each characteristic, only bridges with homogeneous structural characteristics were selected from the concrete highway bridge data set. For example, a bridge with a seat type abutment (non-monolithic), and a diaphragm type (monolithic), was defined as a heterogeneous bridge and excluded from the homogeneous data set. Similarly, a bridge with both multiple and single column bents was defined as a heterogeneous bridge, and was excluded from the homogeneous data set. Bridges with incomplete information were also excluded.
Another criterion in the data selection relates to the correlation analyses. A complete data set for correlation analyses requires that all the bridges exposed to a given ground shaking level be included. In order to satisfy this requirement, a minimum PGA level was selected as a threshold value. This PGA level was determined based on the available ground motion maps. The data set that satisfies this condition was extracted from the homogeneous data set and is referred to as the correlation data set.
The bridges in the correlation data set were grouped first by the superstructure type and substructure material. Then, these bridges were further classified into subcategories based on other structural characteristics, such as number of spans, abutment type, column bent type and span continuity. The classification scheme used in this task was adapted from the bridge classification developed by Basöz and Kiremidjian (1996) under another NCEER project. The damaged bridges were classified to group bridges together that were expected to experience similar damage levels under a given seismic loading. The correlation studies were also carried out using the bridge classification defined by the National Institute of Building Sciences (NIBS) Manual (RMS, 1995).
Reliability of the Database: Caltrans currently has two database systems: the first one (SMS) follows the Federal Highway Administration (FHWA) National Bridge Inventory System but is more detailed, and the second one (BIRIS) is a database that stores bridge books and drawings for all the bridges. The bridge books include detailed reports from each bridge inspection. For some of the 25,000 bridges in the state of California, discrepancies exist between the two databases.
The two databases were compared to verify the correctness of the attribute values for some of the bridges that were damaged in the Northridge earthquake. The abutment type and column bent type were found to be more likely to have errors than the other attributes of interest for this research. For example, in only a few cases (for 2 to 3% of the damaged bridge data set), the values of the design year and skew attributes were found to be incorrect. Where there were discrepancies between the two databases, the structural plans were investigated with the assistance of bridge engineers from Caltrans to determine the correct attribute values. In
order to evaluate the effect of error inherent in the database on the results of the correlation studies, several analyses were performed using sample data sets with corrected and uncorrected data.
Correlation analyses were performed using the data on bridge damage and repair cost. The correlation studies were performed for the following objectives:
- To determine the structural characteristics that best represent damage such that bridges can be grouped using these characteristics,
- To obtain ground motion-damage relationships for bridges with similar structural characteristics,
- To obtain ground motion-repair cost ratio relationships to estimate direct economic losses due to damage to bridges,
- To correlate damage and repair cost ratio.
The data on bridge damage were compiled in the form of damage matrices, i.e., the number of bridges with each level of observed damage at different PGA levels. Then, the damage probability matrices (DPMs), i.e., the probability of being in a damage state given the ground motion level were obtained for each group of bridges. The damage matrices were used as input data to logistic regression analysis to obtain empirical fragility curves both unconditional and conditional on damage. Similar procedures were used to obtain empirical fragility curves for the repair cost ratio. Comparison of observed damage data to currently available ground motion-damage relationships (ATC, 1985; RMS, 1995) were presented in Basöz et al. (1997).
Example Statistics and Results
Some of the results based on data from the Northridge earthquake are presented in this section. The bridges in the Greater Los Angeles area including Los Angeles, Ventura, Riverside, and Orange Counties, were exposed to ground shaking during the 1994 Northridge earthquake. Table 1 lists the number of state and local bridges and the number of damaged state bridges in each of the four counties. A database that includes state and local bridges for the four counties was extracted from the Bridge Maintenance Database compiled by Caltrans (1993). Structural characteristics such as structural type and material, number of spans, abutment type, span continuity, design year indicating the seat width and column longitudinal reinforcement, substructure type, skew, and foundation type were included in this database.
Table 1: Distribution of State and Local Bridges and the Number of Damaged State Bridges in the Greater Los Angeles Area County No. of State Bridges No. of Local Bridges Total No. of Bridges No. of Damaged Bridges Los Angeles 2,097 1,553 3,650 228 Riverside 644 338 982 - Orange 463 505 968 - Ventura 329 175 504 5 Total 3,533 2,571 6,104 233
Bridge damage from the Northridge earthquake pertained mostly to state bridges in Los Angeles and Ventura Counties. Bridges in these two counties also experienced much higher accelerations than those in Riverside and Orange Counties. A total of 63 bridges were exposed to peak ground acceleration (PGA) levels of 0.15g or higher in Riverside and Orange Counties. As shown in Table 1, 3,533 state and 2,571 local bridges are located in the four counties. Of the 3,533 state bridges, 3,318 (1,902 bridges in Los Angeles County, 312 in Ventura County, 462 in Orange County and 642 in Riverside County) carry highway traffic and were included in the highway bridge data set. The number of bridges in the highway bridge data set by superstructure type and substructure material are shown in Table 2.
Table 2: Distribution of Highway Bridges in Los Angeles, Ventura, Orange and Riverside Counties by Structural and Material Type Concrete Steel N/A Timber Concrete Girder 2,396 91 708 0 Steel Girder 119 3 32 0 Truss 4 0 0 0 Tunnel 0 0 4 0 Timber 0 1 0 4 Arch 12 0 15 0 Suspension 1 0 0 0 Unknown 4 0 6 0 1 These nine bridges have concrete slab type superstructure and have both concrete and steel column bents.
Figure 1 shows the distribution of these bridges by design year. Seventy seven percent of the bridges in the highway bridge data set were designed by pre-1971 design standards. The majority of bridges in the four counties were concrete structures (see Table 2) as were more than 85% of the damaged bridges. Therefore, the statistical analyses were conducted mainly for concrete bridges. Figure 2 shows the recorded PGA values (USGS, 1994) and the bridges in Los Angeles, Ventura, Orange and Riverside counties. The PGA value at a given bridge site was obtained within GIS.
Figure 3 shows the distribution of damaged bridges by design year and damage state. Seventy seven percent of the damaged bridges were designed with pre-1971 design standards. Eighty percent of the damaged bridges were multiple span bridges. Figures 4 and 5 show fragility curves for multiple span bridges. Note that the empirical fragility curves shown in Figure 5 are conditional on damage, i.e., they show the probability of being in (or exceeding) a particular damage state given a bridge is damaged. The PGA values shown on the horizontal axis are the observed values reported by USGS (1994).
Table 3: Distribution of Estimated Repair Cost by Damage State Damage State
Number of Bridges
Estimated Repair Cost
For the repair cost data, a database was compiled from the supplementary bridge damage reports provided by Caltrans. A total of about $150,000,000 was reported as repair cost in these reports. The total repair cost for the six collapsed bridges constitute 75% of the repair cost of all damaged bridges. The database includes total estimated repair cost and more detailed information on repair work and cost for 130 bridges in Los Angeles and Ventura counties. Table 3 shows the estimated repair cost of damaged bridges by damage state. Figure 6 shows fragility curves for repair cost ratio for multiple span bridges.
The results from this task can be used to assist in making decisions about mitigation, such as prioritizating bridges for seismic retrofitting, and for post-earthquake response and recovery activities. More specifically, the areas that can benefit from the results of this task include the following:
Database of bridge damage and structural characteristics: A comprehensive database will be available for the most recent two major earthquakes in the United States. This database includes information on bridge damage and the structural characteristics that are important for vulnerability assessment of bridges. This type of a database provides an essential base to improve our understanding of bridge damage from past earthquakes.
Classification of bridges and ground motion-damage relationships: Currently available bridge classes and the corresponding ground motion-damage relationships are rudimentary and do not properly estimate the observed damage from the Northridge earthquake. The method used in this research utilizes the observed damage data to develop empirical fragility curves which can and should be improved as more data become available. PGA levels, skew, span continuity, abutment type, and number of spans correlated well with observed damage. These are the characteristics used in the classification by Basöz and Kiremidjian (1996) and suggests that data shows good agreement with the structural characteristics used in that bridge classification.
Damage state definitions: Post-earthquake damage assessment is an important area that needs to be addressed for efficient and effective emergency response management. The damage states proposed in this research can be used to develop a post-earthquake investigation form that will assist in compiling bridge damage.
ATC-13, (1985), "Earthquake Damage Evaluation Data for California," Report ATC-13, Applied Technology Council, Redwood City, California.
Basöz, N., Kiremidjian, A.S., King, S.A. and Law, K.H.. (1997), "Characteristics of Bridge Damage in the 1994 Northridge, CA Earthquake," submitted to Earthquake Spectra.
Basöz, N. (1996), "Risk Assessment for Highway Transportation Systems," Ph.D. Dissertation, Department of Civil Engineering, Stanford University, (July).
Basöz, N., and Kiremidjian, A.S., (1996), "Prioritization of Bridges for Seismic Retrofitting," Technical Report No. 118, John A. Blume Earthquake Engineering Center, Civil Engineering Department, Stanford University, Stanford, California.
Buckle, I.G. (1994), "The Northridge, California Earthquake of January 17, 1994: Performance of Highway Bridges," Technical Report NCEER-94-0008.
California Department of Transportation (Caltrans), (1994), "The Northridge Earthquake," Caltrans PEQIT Report, Division of Structures, Sacramento, CA.
California Department of Transportation (Caltrans), (1993), "OSM&I Coding Guide for SMS," Division of Structures, Office of Structures Maintenance and Investigations, Sacramento, California.
California Department of Transportation (Caltrans), (1989), "The Loma Prieta Earthquake," Caltrans PEQIT Report, Division of Structures, Sacramento, CA.
EERI, (1995), "Northridge Earthquake of January 17, 1994 Reconnaissance Report," Earthquake Spectra, Earthquake Engineering Research Institute, Oakland, California.
Risk Management Solutions (RMS), (1995), "Development of a Standardized Earthquake Loss Estimation Methodology," Prepared for the National Institute of Building Sciences by Risk Management Solutions, Inc., Menlo Park, California.
Somerville, P., Saikia, C., Wald, D. and Graves, R., (1996), "Implications of the Northridge Earthquake for Strong Ground Motions from Thrust Faults," Bulletin of the Seismological Society of America, Vol. 86, No. 1B, (February), S115-S125.
USGS, (1994), US Geological Survey, Open-File Report 94-197, Menlo Park, California.
Woodward-Clyde Federal Services (WCFS), (1995), Contoured Ground Motion Parameters for the 1994 Northridge Event, ASCII files.
Yashinsky, M., Hipley, P. and Nguyen, Q., (1995), "The Performance of Bridge Seismic Retrofits During the Northridge Earthquake," Caltrans Office of Earthquake Engineering, Sacramento, California.
NCEER Bulletin, April 1997, Vol. 11, No. 2