Benchmarking of FEMA P-58 Expected Seismic Losses to Observed Loss Data from the 1994 Northridge Earthquake

By Dustin Cook, P.E.

Recent developments in seismic risk analysis and performance-based design have opened doors to statistically rigorous and systematic frameworks for the quantification of seismic risk to buildings. One such method was recently formalized by the Applied Technology Council and is known as FEMA P-58: Seismic Performance Assessment of Buildings (FEMA, 2018c). FEMA P-58 provides a probabilistic approach to assess the seismic risk of individual buildings. Through FEMA P-58, seismic risk is quantified in terms of performance metrics that are valuable to engineers, building owners, and the public, such as economic losses, potential casualties, and disaster recovery time, facilitating decisions among design or mitigation alternatives. One of the pressing challenges of new and evolving seismic risk assessment methods is the need to validate, confirm, and/or verify the outcome of these assessments to support their broader use.

Records and estimations of damage and loss from the 1994 Northridge earthquake in southern California provide a unique opportunity to benchmark new performance-based risk assessment methodologies, such as FEMA P-58, to empirical data, especially for buildings of wood frame construction. To evaluate the results obtained through a FEMA P-58 assessment, this study compares the hindcast losses using the FEMA P-58 methodology, in terms of repair costs of buildings, with observed losses during the 1994 Northridge Earthquake. This article summarizes the methods and findings of recent a study performed jointly by the University of Colorado Boulder and the Haselton Baker Risk Group.

Figure 1 – Highway damage during the 1994 Northridge Earthquake. Photo from David Butow/Corbis/Getty Images.

Unprecedented levels of damage were observed during the Northridge Earthquake, with total estimated losses ranging between $40 and $44 billion (1994 USD), making it the costliest earthquake in U.S. history (Eguchi, 1998). Other sources estimate an additional $8 billion in indirect losses from business interruption, lost tax revenue, vacated housing, and defaults on Small Business Administration loans (Petak & Elahi, 2001). Widespread damage was observed in many different types of buildings across a large region. Wood frame structures make up a majority of the building stock in the US, and represent around 96% of the buildings in Los Angeles County, accounting for around 85% of the value of the building stock at the time of the Northridge Earthquake. Damage to wood frame structures was extensive due to the large number of buildings that were affected. In total, over 340,000 insurance claims were submitted for damage to residential structures, mostly from damage to nonstructural components (Eguchi, 1998). Considering both insured and uninsured damage, losses from wood frame residential structures are estimated to represent about half of the total building losses from the earthquake (Petak & Elahi, 2001).

FEMA P-58 only quantifies direct losses to the building (i.e. structural and nonstructural components), and does not represent losses to contents, detached structures, and loss of use. Therefore, adjustments are made to the estimated total direct losses to derive a best estimate of direct loss from building damage, to compare against FEMA P-58 losses.  In the Northridge Earthquake, direct building loss represented about 67% of the total loss from insurance claims (Eguchi, 1998, Petak & Elahi, 2001). We propose the reduction factor to go from reported loss to losses comparable to FEMA P-58 is likely to be somewhere between 60% and 75%, resulting in a range of estimated loss from $25 billion to $32 billion by multiplying the average reported estimated direct loss of $40 to $44 billion (Petak & Elahi, 2001) with this ratio range.

To quantify FEMA P-58 expected loss from the Northridge earthquake, this study uses the Haselton Baker Risk Group’s FEMA P-58 Risk Model (SP3-RiskModel) and the USGS Northridge ShakeMap to perform a scenario assessment on a set of 2.6 million buildings affected by the event. The SP3-RiskModel uses embedded algorithms to generate structural properties, building configurations, and building components from a simplified set of building inputs based on typical building configurations and inventories, engineering analysis, and expert judgment. This tool helps expedite the FEMA P-58 analyses for large building inventories with limited building information, such as the building data available in tax assessor databases, and can be used to run millions of performance models in a batch setting. This study uses a proprietary inventory database based on tax assessors and insurance data that contains building inventories of the greater Los Angeles region. Losses from the assessment are aggregated for all buildings within the ShakeMap region and compared with observed data.

Table 1 – Breakdown of FEMA P-58 predicted loss for each type of building where quantitative loss data is available. Dollar values represent 1994 USD.

Type of Building Number of Buildings Total Value FEMA P-58 Predicted Loss          Estimated Observed Loss
All Buildings 2,586,638 $722 billion $31 billion $25 to $32 billion
Residential 2,405,382 $508 billion $16 billion $12 to $16 billion
Tilt-Ups 39,677 $52 billion $5 billion > $1 billion


Figure 2 – 1994 Northridge Earthquake mean losses from FEMA P-58 for part of Los Angeles County.

The total hindcast losses from the FEMA P-58 assessment of the Northridge scenario came out to $31 billion due to building damage, which is on the higher end, but well within the range of observed loss. Mean loss results presented here represent the average cost, in 1994 USD, to repair structural and nonstructural components in the building. The total hindcast loss from residential buildings also compare well with observed losses. This comparison indicates that probabilistic methods such as FEMA P-58 can provide accurate predictions of post-earthquake economic losses.  However, results from probabilistic methods are heavily dependent upon modeling decisions. If factors that affect the response of structures at low levels of shaking are not properly accounted for, such as effective damping and stiffness, predictions of loss can change significantly. Outcomes of alternative modeling decisions are discussed in the full report. For more information on the complete findings of the study, please contact the Haselton Baker Risk Group at (530) 531-0295 or

Dustin Cook is a Ph.D. Candidate in the Department of Civil, Architectural, and Environment Engineering at the University of Colorado Boulder and is a member of the Haselton Baker Risk Group. Dustin holds a M.S. from University of California, Los Angeles and is a licensed Professional Engineer in the state of California. You can reach him by email at


Younger Members Committee