Earthquake hazard maps may overestimate shaking dangers

Scientists working on the California Historical Intensity Mapping Project compare quake-shaking forecasts to measured shaking intensity, with some surprising results.
 

By Julie Pierce Onos (@JuliePierceOnos)
 

Citation: Pierce Onos, J. 2020, Earthquake hazard maps may overestimate shaking dangers, Temblor, http://doi.org/10.32858/temblor.128
 

Scientists test a hypothesis, or prediction, by collecting data and comparing them to the hypothesis. Testing hypotheses about earthquakes, which are infrequent, often involves using data from past earthquakes — a retrospective approach to figuring out what might happen in the future.

In a new study, published in Seismological Research Letters, scientists used this retrospective approach, or hindcasting — the opposite of forecasting — to explore how well earthquake hazard maps forecast maximum ground shaking. Data from past earthquakes suggest that the current maps may overestimate future shaking.

 

Reinterpreting historical seismic intensity data

In the U.S., the U.S. Geological Survey (USGS) puts together National Seismic Hazard Maps to estimate the maximum shaking an area might be subject to over a defined period of time. Such hazard maps are used to produce local earthquake-resilient building standards, affecting millions of people and billions of dollars in construction costs worldwide, so it’s important to get them as accurate as possible.

The field of hazard mapping was taught an important lesson by the March 2011 Tohoku, Japan, earthquake and tsunami that killed more than 20,000 people. The hazard maps indicated that a magnitude 8.0 was the maximum magnitude quake to which the area would be subject. However, a magnitude-9.0 earthquake rocked Tohoku and triggered a devastating tsunami. Japan’s hazard maps had shown Tohoku as a lower hazard area than it turned out to be.

Inspired by the Tohoku earthquake and tsunami, Seth Stein, a seismologist at Northwestern University, Susan Hough of USGS and a team of graduate students at Northwestern launched a project to explore how well hazard models predict future shaking. The team’s approach, Stein says, was “like the way weather forecasts are evaluated to assess how well their predictions matched what actually occurred.”

Initial research focused on Italy from 217 B.C.E. to 2002 C.E. The team compared the probable shaking predicted by Italy’s hazard maps to historical accounts of shaking from earthquakes. The level of shaking is described by a number on the Modified Mercalli Intensity (MMI) scale. This goes from 1 (shaking not felt) to 10 (extreme). The team found the intensity of the shaking at various locations was much lower than indicated by the hazard map. The same result arose from analyzing a 510-year-long record of Japan’s seismic activity: The intensity of shaking was less than what was anticipated.

 

Compiling the data

These findings inspired Stein and colleagues to investigate how well California’s hazard maps forecast hazard, as measured by the intensity of shaking from past earthquakes. The team created the California Historical Intensity Mapping Project (CHIMP) dataset, which they’ve made publicly available for any researchers to use to learn more about past seismic activity in California.

 

 

The CHIMP dataset provides the maximum shaking at sites in the past 162 years. Creating the dataset involved three components: USGS’s “Did You Feel It?” (DYFI) data, historical accounts and personal accounts. For recent quakes, the team used shaking intensity values from DYFI, which takes citizens’ accounts of shaking reported over the internet and assigns location-based intensity values using an algorithm. Because the DYFI data are only from the last 20 years, the team turned to historical accounts to look further back in time. They used previously compiled historical accounts — like newspaper reports about people’s descriptions of a window breaking or what fell off the wall — and collected some oral histories.

Northwestern doctoral students Leah Salditch and Molly Gallahue traveled to the Southern California desert to conduct in-person interviews to collect oral histories from long-time residents who recalled the 1993 magnitude-6.1 Big Pine earthquake and 1992 magnitude-6.1 Joshua Tree earthquake. The team assigned shaking intensity values to each account to produce data like the DYFI data. Producing intensity data from historical accounts like this is “a powerful tool,” Stein says.

 

 

Gallahue says one interview really stood out. A woman arrived with newspaper clippings and “a whole photo album because her house was hit so hard by this earthquake — thankfully, no one was injured. But she just had so many stories and was so excited to tell us because everyone called it ‘Lupe’s earthquake’” — Lupe is the woman’s name.

It’s interesting to consider how different interviewees’ stories were based on their level of preparedness for the quake, Salditch says. “Some people were completely caught off guard and had no idea what to do. Some people were so prepared that they experienced much less shaking than anybody else,” Salditch says. That reveals one challenge in relying on “felt reports” — everyone experiences things differently. One woman who was interviewed told the researchers how her husband, who was fascinated with emergency preparedness, had attached every single piece of furniture and bookcase to the walls so she was only aware of the earthquake because of the commotion occurring when her neighbors were rushing outside. Salditch and Gallahue conducted these interviews to glean both the accounts of felt shaking and to collect information regarding the damages. Both types of information were used to inform the intensity values assigned in the dataset.

 

 

Building a shaking intensity database

The USGS’s hazard map for California is based on fault rupture forecasts from the Uniform California Earthquake Rupture Forecast, Version 3 (UCERF-3), and ground-motion models from the Next Generation Attenuation-West2 Project. The process does not explicitly incorporate past shaking intensity data, Stein notes. The CHIMP dataset does, which makes it uniquely valuable. CHIMP contains 46,502 intensity observations from 62 earthquakes that occurred during the past 162 years with estimated magnitudes ranging from 4.7 to 7.9, Stein says. The team compared the CHIMP dataset with what was predicted by the California hazard map and found “these [intensity observations] show maximum shaking throughout California lower than expected from seismic hazard models,” he says.

 

 

That doesn’t necessarily mean that current forecasting and ground-motion models require revisions to be consistent with the new shaking intensity data, however, Stein notes. “Possible reasons for the discrepancy between the hazard maps and CHIMP include: limitations of the CHIMP dataset, biases in the hazard models, and the possibility that California seismicity throughout the historical period has been lower than the long-term average used in UCERF.”

 

The future of earthquake shaking intensity maps

How and when to incorporate CHIMP’s findings in California’s hazard map is an open question. But CHIMP has other applications as well. One is revisiting older earthquakes’ magnitudes and locations in California. In the early 1900s, seismic stations were neither widely distributed nor highly sensitive. CHIMP data could help refine the locations and magnitudes of these past quakes, says Norman Abrahamson, a geophysicist at the University of California, Berkeley who was not involved in this research. Moreover, he notes, these data could help explore the distribution of recent and historic earthquakes’ ground motions.

Abrahamson is currently using the CHIMP database to analyze regional differences in shaking from large-magnitude earthquakes. According to Abrahamson, CHIMP is significant because it “is the only dataset of large-magnitude earthquakes that has enough earthquakes to do this test.” Also notable, he says, is how well the CHIMP project developed a useful methodology for evaluating qualitative data such as eyewitness reports of felt shaking and damage into information that can be used for estimating seismic intensities.

 

 

Salditch says she hopes “that the historical earthquake community and the earthquake early warning community can come together and make seismic intensity a more robust or trusted measurement.” One way she envisions this being done is by merging the CHIMP dataset with future DYFI data since both datasets are made up of intensity data. Work would have to be done to ensure consistent interpretations, she notes, since DFYI takes citizens’ accounts of shaking assigns intensity values using a computer algorithm with its own parameters, whereas CHIMP used different methodology for assigning intensity values.

The CHIMP team continues to examine how the dataset may change over time and is working to eliminate biases in how they interpret the data even as they add to it. They are also developing better methods to compare the observed shaking to hazard maps. Achieving their goal of improving earthquake hazard maps will allow those who rely on them to better prepare for and mitigate earthquake damage and loss.

 

Further Reading

Salditch, L., M. M. Gallahue, M. C. Lucas, J. S. Neely, S. E. Hough, and S. Stein (2020). California Historical Intensity Mapping Project (CHIMP): A Consistently Reinterpreted Dataset of Seismic Intensities for the Past 162 Yr and Implications for Seismic Hazard Maps, Seismol. Res. Lett. XX, 1–20, doi: 10.1785/0220200065.

Stein, S., Brooks, E., Spencer, B., Vanneste, K., Camelbeek, T., Vleminckx, B. (2017). Assessing how well earthquake hazard maps work: Insights from weather and baseball. Earth Magazine.