The Resilience Divide Part 1: Inequity in Seismic Impact

Research by Abhineet Gupta, Anne Wein, and Jon Haveman

Disasters do not only expose inequalities — they amplify them. Before, during and after a disaster, data increasingly demonstrates that demographic factors play a large role in determining one’s resilience to disaster events.

Businesses are not immune to these differences in outcomes. Certain factors or characteristics of businesses — such as age, ownership or industry — may be related to their vulnerability during a disaster.

In the first installment of our Resilience Divide series, we will be exploring our own research in this area, which builds upon previous research done at the USGS. By adding our own findings to the conversation, we hope that future studies can shed further light on these issues — helping to uncover the underlying patterns behind these disparities in disaster resilience.

Image for post
Image for post
The team presented this project at the AGU Fall Meeting in 2019. For more information, please visit

Seismic Vulnerability

With regard to earthquakes, a number of researchers are interested in analyzing the discrepancies in impacts across businesses, and how business characteristics may have a relationship to seismic vulnerability. In 2019, research at the USGS analyzed these impacts for the HayWired scenario. Used for a variety of similar studies and analyses, the HayWired scenario is an impact simulation that anticipates damage from a magnitude 7.05 (M 7.05) earthquake on the Hayward fault in the California Bay Area.

Wein, Haveman et al.’s analysis demonstrated that location and sector (or industry) of a business are strong differentiators of disruptive building damage risk, and that minority ownership, branch ownership structure, and low revenue are other above average differentiators. The analysis used building data at the census tract level, which encapsulates an area about the size of a neighborhood, or 1200–8000 people.

At One Concern, we were interested in collaborating on these efforts. As a company, and as scientists, we are invested in understanding the factors and nuances that contribute to overall disaster resilience, and how different factors like business or population demographics affect the resilience of communities.

In 2019, we collaborated with Anne Wein from USGS and Jon Haveman from Marin Consulting to expand on their HayWired analysis — incorporating multiple earthquake scenarios (rather than one), and using building-level data in lieu of census tract level data that One Concern has compiled from various sources using machine learning. By expanding on their study, we aimed to provide a more nuanced, high-resolution analysis of damage impacts, in addition to analyzing how impacts differ across earthquakes, regions and communities.


Ground Motion Generation

In the Bay Area, there are two large faults that have a high likelihood of a magnitude 7.0 or greater earthquake in the next 30 years: the San Andreas fault, which runs along the Peninsula and near San Francisco, and the Hayward fault, which runs through the East and North parts of the bay.

Estimating ground motions from earthquake ruptures is one of the first steps to understand their impacts in a region. Ground motion — the result of seismic waves — refers to the measurement of how much the earth’s surface moves at every location in a region during an earthquake. Measuring acceleration and velocity of the movement, ground motion is a critical factor in determining the expected probability of earthquake-related damage.

To obtain a broader picture of risk, and gain insight into the different levels of damage for buildings across the region, the One Concern team expanded on the prior analysis of the HayWired scenario. To start, we generated two new sets of ground motions corresponding to two earthquake ruptures using the OpenQuake engine — an M7.0 earthquake on the Hayward fault and an M7.0 earthquake on the San Andreas fault. While the earthquake magnitude on the Hayward fault is similar to that of the HayWired scenario, difference in location of the earthquake epicenter and inherent randomness in wave propagation provided variations in the ground motions.

The three ground motion generations are shown below:

Image for post
Image for post
M7.05 HayWired Scenario, used in the prior study
Image for post
Image for post
M7.0 San Andreas
Image for post
Image for post
M7.0 Hayward

Exposure and Vulnerability Functions

Next, in order to estimate how businesses would be impacted in each of the earthquake scenarios, the team had to assess building characteristics, and eventually building damages, for each business in the Bay Area. For this data, the NETS database provides business characteristics like business location, sector, and ownership; however, it does not include building characteristics like construction year and material, which are required to estimate the buildings’ vulnerability during earthquakes. One Concern had compiled a database of individual buildings by combining multiple datasets like the census and tax assessor’s data using machine learning tools, which we then combined with the NETS database to associate relevant building characteristics to each business location.

This step provided further granularity to the previous study, which used aggregated building characteristics at a census tract level based on occupancy class to estimate predicted damage level to a given business. Our dataset allowed us to identify individual business characteristics — like construction material and year — which were used to estimate the damage to the buildings in the generated earthquake scenarios by associating fragility functions to each building.

Building Damage Estimation

HAZUS fragility functions (or fragility curves) assess predicted damage to a building by associating probability of damage to the building given different ground motion levels. The curve represents predicted damage level as a probability distribution between four states: Slight, Moderate, Extensive, and Complete. Each curve represents the probability of experiencing that level or greater damage.

Image for post
Image for post
Example of a HAZUS fragility curve, from Hazus-MH 2.1 AEBM User’s Manual

Taking into account the building type and code level based on its construction year for each individual building, the team assigned HAZUS fragility functions to the buildings. Similar to the prior HayWired analysis, the team focused on the fragility curve corresponding to the extensive damage state, or the probability that the building experienced “extensive” or “complete” damage. By combining the ground motions with the fragility curve, we could estimate the probability of extensive or greater damage for each of the scenarios, as shown below.

In order to provide a comparison with the results from the previous study, we also estimated the damage for the HayWired scenario. Ground motions are the same as in the previous study; however, the building characteristics and fragility functions are based on One Concern models.

Image for post
Image for post
The probability of extensive or complete damage at business locations in each of the three scenarios


Comparison of HayWired Results with Prior Study

As previously stated, the prior study on the HayWired Scenario used census tract level HAZUS data, while our study used individual building characteristics data. To verify this analysis, we compared the results for the HayWired scenario from our study with the prior study.

We observed an increase in building damage, with the probability of extensive or complete damage for the entire Bay Area increasing to 11.7% from 7.3% in the prior study. This increase is within a reasonable range given the change from aggregated census-level building data to individual building data.

In the prior study, the building damage probability was an average probability of damage for all buildings in the same occupancy category in a census tract; while in this study, if the business building was more vulnerable than the average building, we would expect a higher probability of damage. However, in certain cases — for example, in San Francisco — we observed an increase in probability of damage, from 1.3% in the prior study to 15% in this study. This is a considerable change, and initial spot checks indicated that the increase occurred due to buildings in some sectors like Utilities being associated with old unreinforced masonry buildings that are highly vulnerable to earthquakes. A more in-depth review is needed to further identify the factors contributing to this increase in damage, which was not done within the scope of this study.

Business Characteristics Associated with Building Vulnerability

For each of the three scenarios — HayWired scenario, M7.0 earthquake on Hayward fault and M7.0 earthquake on San Andreas fault — we grouped the businesses by their characteristics like location, business sector, and ownership to identify the characteristics that are at higher risk of building damage during these earthquakes.

Across all three scenarios, the following trends were observed:

As business age, revenue, and size increased — in other words, the older, more profitable, or bigger the business — the percentage of buildings with extensive or complete damage increased. Accommodation and food sectors were among the highest impacted across all scenarios, and minority-owned businesses, on average, experienced more extensive damage than those that were not minority-owned.

However, there were other trends that did not carry across all scenarios — for the two fault lines studied, the team found slightly different impact outcomes. For the two Hayward fault scenarios, manufacturing and public administration were among the most impacted, while information and utilities sectors were most impacted in the San Andreas scenario. This is likely a result of business locations, as the San Andreas scenario has a higher impact in San Francisco and the peninsula regions where more information sector businesses are located.

In addition to estimating the total percentage of buildings with extensive or complete damage, we also estimated the employee-weighted percentage of buildings with extensive or complete damage, in order to understand how employees would be impacted based on different business characteristics. Across all three scenarios, we weighted the risk by the number of employees — observing elevated risk for businesses (compared to risk without weighing by employees) in Santa Clara County, for the manufacturing, public administration, wholesale, and utilities sectors, as well as for female-owned businesses. However, this weighing maintained the trend for size, revenue, and minority-owned business characteristics.

Conclusions and Limitations

While the business sector proved to be a strong indicator for building damage, and accommodation and food services were among the most impacted sectors across all scenarios, the team found that business location — specifically, the county of business location — was the strongest differentiator of building damage.

Manufacturing and public administration sectors were found to be more highly impacted in Hayward fault scenarios, while the information and utilities sectors were more impacted in the San Andreas scenario. This indicates that in order to more accurately understand the vulnerability of businesses based on their characteristics, we need to assess them across multiple earthquake scenarios.

We note that more work is needed to explain the differences in the results for the Haywired scenario between the prior study and this study. Additionally, neither of the studies have validated building characteristics against on-the-ground data, which would also help validate the results.

We believe that this approach and future studies could be used to inform data-driven policies around business and community resilience. By highlighting discrepancies in disaster risk, we can make connections between these differences in impact and larger social patterns, exposing inequities across factors such as race and socioeconomic status. Building resilience — for all areas of society — cannot be done without understanding the extent to which these inequities are felt.

By identifying business characteristics that are more vulnerable to disruption or closure in a disaster, we can begin to investigate the reasons behind those vulnerabilities — potentially identifying larger patterns within city planning or community investment. From there, we can identify areas of improvement and address the issues head-on.

With these insights, we hope to continue contributing to a much larger conversation around data, disaster mitigation, and above all — building resilience in communities, everywhere.

If you enjoyed this installment of “The Resilience Divide,” stay tuned for upcoming pieces in the following weeks!

Written by

We’re advancing science and technology to build global resilience, and make disasters less disastrous

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store