Is It Possible to Predict “Secondary Damage” from Earthquakes, Typhoons, Floods, and Other Disasters?

Junichi Sakai, Flood and Seismic Engineering Lead

I was given an opportunity to write this article to mark the 10th anniversary of the Great East Japan Earthquake (2011 Tohoku Earthquake). I saw damage caused by the Great Hanshin earthquake in 1995 through news reports when I was a college student, and this inspired me to become a researcher in earthquake engineering. More than 25 years have passed since then. I spent 18 years as a researcher in earthquake engineering for bridges, and thereafter worked as an engineer developing probabilistic risk models for assessing natural catastrophe risks. During these years, I visited damaged cities and towns for reconnaissance surveys after a number of disasters, including the 2011 Great East Japan Earthquake, severe floods in the Kanto region in 2015, the 2016 Kumamoto earthquakes, as well as severe wind and flood events due to Typhoon Faxai and Typhoon Hagibis in 2019. At present, I am engaged in the development of models[1] for assessing disaster resilience by integrating disaster science with AI and machine learning technologies at One Concern, a U.S. Silicon Valley-based disaster prevention start-up.

Disaster resilience means disaster responsiveness. In general, the concept of resilience referred to the ability to deal with external impacts. However, since the 2011 Great East Japan Earthquake, the term has been widely used in disaster-related contexts as the ability to endure an unprecedented situation, recover from it, and prepare in advance.

Attention being paid to “indirect damage” from disasters

Once a big natural catastrophe occurs, people’s attention tends to go to direct damage, such as loss of human lives and damage to buildings or infrastructures. However, indirect damage caused by disruption of electricity, water, gas supply, and transportation infrastructure also has an impact on a company’s business continuity and thus regional or national economy consequently, and is gaining attention recently. In particular, the Japanese government is now promoting “initiatives for building national resilience and preventing and mitigating disasters”[DW1] [2]. In its three-year emergency response plan starting in 2018, the government specifically mentions two perspectives of focus: maintaining critical infrastructures supporting disaster management to save people’s lives from direct damage, and maintaining the functions of critical infrastructures supporting the national economy and the everyday lives of the people (e.g. electricity and water services). This modernized focus illustrates how important it is to maintain the functions of lifeline facilities and infrastructure after disasters.

These initiatives taken by the Japanese government are mainly focused on tangible measures like enhancement of performance of buildings/infrastructures, e.g. seismic retrofit of infrastructures or enhancement of river levees, , and they are expected to reduce risks, loss of human lives, and direct damage to buildings or infrastructures , and also to enhance the disaster resilience of lifeline facilities. However, it will take some time before it bears fruit. In the meanwhile, it is necessary to consider intangible measures, such as understanding the current level of disaster resilience accurately with the use of the latest technologies and preparing mitigation strategies so that these disaster risks can be prevented.

In Japan, where natural disasters occur almost every year, many people understand that there will be loss of human lives and damage to buildings from disaster events. Meanwhile, with regard to indirect damage, many people are probably still not sure how the direct impacts of a disaster will give way to a ripple effect of indirect damages. This is because indirect damage spread depends on complicated conditions intertwined with even more complicated interrelations. At this stage in our path towards resilience, the first step is to visualize hidden disaster risks.

Difficult to estimate hidden disaster risks

Technology that estimates damage to buildings and infrastructure caused by earthquakes, typhoons, floods, and other natural disasters has a relatively long history (around 50 years at the longest, though), and the technology has evolved with the evolution of computer technology. In recent years,some researchers have been working on a research project that conducts an earthquake simulation for an entire city using a supercomputer[3]. Their main focus has been simulation for direct damage from earthquake shaking, but due to the uncertainty of ground shaking and dynamic response of structures, it is not easy to estimate the damage to each building with a certain level of accuracy.

What is even more difficult is to estimate indirect damage caused by disruption of electricity, water, gas supply, and transportation infrastructure. This is because you first must be able to estimate direct damage to buildings and other structures, as well as lifeline facilities, with a certain level of accuracy. Then, you need to further analyze the impact of the disruption of lifeline networks due to the damage to the lifeline structures..

For example, in analyzing impact on a company’s supply chain, you would first examine the connection of routes (such as parts procurement and product supply routes) used by the company. You would then consider various factors, such as the location of each site, the extent of direct damage from a natural disaster, the time needed for the sites to restore their normal operations, and the presence or absence of alternative sites that can supply the necessary parts and products.

Taking a manufacturer as an example, it is necessary to estimate not only direct damage to and restoration of plant buildings and facilities, but also that of supply resources, such as electricity, water, and gas to the plant, in addition to roads, railways, ports, airports, and other transportation infrastructure necessary to transport raw materials and parts. In the case of manufacturers, since a variety of parts may be supplied from various sites, analyses need to be conducted for each site. Naturally, as the number of sites increases, uncertainty will expand. Under such conditions, conducting impact analyses while maintaining a certain level of accuracy is quite challenging.

To overcome such technical challenges and estimate hidden disaster risks, the concept of a “digital twin” (a twin of the real world in the digital realm) is expected to play a key role.

Taking on the challenge of creating a “twin” of the real world

A digital twin refers to a “twin” of a real-world object created in the virtual world for simulation and other purposes. Digital twins have already been put into practical use in the manufacturing industry, and attempts to apply them to cities have recently started. For example, Japan’s Ministry of Land, Infrastructure, Transport and Tourism (MLIT) is working on an initiative named Project PLATEAU[DW2] to reproduce 3D cities in the virtual world. Digital twins applied to cities are expected to reproduce cities in the 3D world for disaster prevention and other purposes, as well as help to realize smart living by utilizing real-time data obtained from sensors and other tools.

One Concern’s digital twin technology specializes in disaster prevention. We started from one city, and then aim to construct a broad area so that we can simulate damage that would occur to the area in a catastrophic event in the virtual world. Although still in the development stage, our goal is to create a digital twin of all of Japan.

If such a twin of the real world can be created, this will make it easier to estimate not only direct damage, but also indirect damage, which has higher uncertainty.

However, there are several technical challenges to overcome in order to minimize uncertainty and increase the reliability of the results from the simulation. Such challenges include not only creating a certain level of damage estimation models for structures, but also collecting the detailed data.

The variety of data required to produce such models is quite challenging. Consider this: A society’s foundation is formed over a long period of time by clearing land for cultivation, creating residential and commercial areas, developing transportation infrastructure such as roads and railroads, creating power plants, and establishing electricity and water services. In addition, river improvement works have been carried out to prevent riverine flooding, while sea walls have been built to mitigate damage from Tsunamis or storm surges. Data on all of these is needed to create a twin of the world, which means you need data on the natural environment, buildings, power grids, water networks, transportation infrastructure — the list is almost endless. (It might be easier to understand if you imagine there are several data layers for each of them shown in the figure.)

In recent years, data is increasingly being organized and some has been made available to the public by governments and other agencies. It is now possible to obtain data related to the natural environment — such as data on elevation, land use, and soil type (VS30)for locations all over Japan. However, it is not easy to obtain data for each building because the estimate of damage from disasters requires detailed data, such as the occupancy type, construction type, shape, materials, year of completion, and number of stories.

The data that is even more difficult to obtain is data on various types of infrastructure facilities owned and managed by a variety of administrators, data related to lifeline networks, as well as data on river levees, sea walls, and others . Among such data, information on the locations of some infrastructure facilities have been made available to the public through the National Land Numerical Information download services by MLIT[[DW3] 4]. However, detailed information on such facilities and data related to networks are not available.

Under these circumstances, expectations are growing for AI and machine learning (ML) technologies that enable missing data to be inputted through algorithms that are developed using available data.

Of course, data predicted in this way may be different from actual data, but we have been trying to secure a certain level of accuracy. As data becomes more organized and openly available, further advances in AI and ML technologies will make it relatively easy to obtain necessary data and create a twin city in the digital world.

Data on damage is the most important, yet difficult to obtain

There is one more important factor from the perspective of data availability: information on damage to and restoration of buildings and lifeline facilities in past disasters is the data that is the most difficult to obtain.

Regarding direct structural damage caused by strong ground shaking or inundation by flood water, much progress has been made in revealing the damage mechanisms through experimental and analytical approaches for many years. In parallel, technology for structure analysis has also advanced. Nevertheless, as there are many things that remain unresolved, one of the most reliable methods is to have the model learn from data on actual damage, which is the ground truth.

Moreover, with regard to recovery, a variety of processes is needed for each subject facility and structure according to their extent of damage, ranging from confirmation of the state of damage to considerations of restoration methods, procurement and exchange of replacement parts, repairing of broken parts, and reinforcement with temporal materials. Current technology has yet to reach a level enabling analytical simulation at this granular level, so the most appropriate method in relation to recovery is also to create a model based on actual cases of post-disaster recovery.

Thus, data on damage to and recovery of buildings and lifeline facilities in past disasters can be the key for the development of a model that assesses resilience, but very little of such data has been made available, as they may be considered confidential information of companies or individuals.

Also, few research institutions, including universities, conduct research for collecting such data or case studies, partly because such research is not likely to be highly evaluated due in part to lack of innovation when compared to research based on state-of-the-art analytical technologies and experiments.

Although the descriptions above seem to indicate we still have a long way to go before becoming able to create a digital twin of the real world and assess resilience, taking the current technological level into account, I believe we have already reached a level where the peak of the mountain can just be barely seen.

However, it is true that we have a steep and long way ahead of us, and the mountain top that we think can be seen may be a mere illusion. Nevertheless, working on the development of state-of-the-art technologies and advancing efforts to commercialize them, step by step, is the path we should take as engineers.

When writing this article, an earthquake that is considered to be an aftershock of the 2011 Great East Japan Earthquake occurred off the coast of the Fukushima Prefecture, and there were news reports about damage to buildings, railway structures of Shinkansen (bullet train), road structures, and supply chains. I was surprised that a major aftershock occurred even after 10 years have passed from the earthquake, but this is a reminder of the fact that as long as we live in Japan, we will always be very close to disasters.

This also made me think again about the extent to which “hidden” disaster risks could be seen in advance. We are exposed to not only aftershocks of the Great East Japan Earthquake, but also risks of the Nankai Trough Mega Earthquake and an inland earthquake occurring directly beneath the Tokyo metropolitan area. Furthermore, large-scale typhoons and localized torrential rains occur almost every year. In order to mitigate the impact from such catastrophic events, the enhancement of performance of buildings and infrastructures conducted by MLIT under the initiative for building national resilience should be promoted.

Since these initiatives cannot be implemented overnight, I hope that the necessity of making risks visible and understanding disaster resilience will be first disseminated widely and a broad range of prior preparations — which are deemed to be intangible measures — will be undertaken based on the results of risk and resilience assessments.

Of course, to do so, it is essential that risks and disaster resilience can be assessed with a certain level of reliability using models, and I hope to contribute toward realizing such models.

In closing, I would like to express my deepest sympathies to those who have suffered from recent disasters.

[1] The model here refers to a series of calculation methods developed to handle natural phenomena and social phenomena quantitatively.

[2] Cabinet Office website: Building National Resilience[DW4]

[3] Muneo Hori: Shake a City as a Whole with a Supercomputer — Simulation of Earthquakes — , Gathering for Learning about Supercomputers in Mito, 2018[DW5] (

[4] The Ministry of Land, Infrastructure, Transport and Tourism: National Land Numerical Information download service

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