Integrate all available data

Takeshi Sagiya

Nagoya University, Japan

If historical records had been more complete, and if discrepancies between data had been picked up, we might have been alert to the danger of a magnitude-9 earthquake hitting Tohoku, even though such an event was not foreseen by the Japanese government.

Rescue workers search for survivors in the village of Nodamura in Iwate Prefecture, northern Japan. Credit: KYODO/REUTERS

In 2002, the Headquarters for Earthquake Research Promotion of the Japanese government released a long-term evaluation of the likelihood of subduction-zone earthquakes in the Tohoku region. It estimated an 80–90% probability that the area would have a large earthquake of magnitude 7.7–8.2 in the next 30 years. But the probability of a magnitude-9 earthquake affecting a 400–500-kilometre area was not specifically mentioned. As a member of the working group involved in the evaluation, it is with great regret that I reflect on the causes of this failure.

Online collection

The long-term evaluation was based on the statistical analysis of the complete historical earthquake record for the past 400 years. However, the Tohoku earthquake clearly shows that 400 years is too short a time period to evaluate seismic activity. In fact, during the past five years, geologists of the Geological Survey of Japan have reported that a great tsunami inundated the coast of the Sendai area in AD 869. It was probably of comparable size to the tsunami that hit after the March earthquake. Unfortunately, those findings came too late to be considered in the evaluation of, and countermeasures against, tsunamis.

One important lesson from this experience is that if we take an empirical approach to evaluating or forecasting natural disasters, all the available information should be taken into account — even though some records have large uncertainties — and all the possibilities should be considered, regardless of their likelihood.

Another weakness of the long-term evaluation was clear discrepancies between the different observational data that went into it. In the past decade, Global Positioning System (GPS) investigations have indicated that the plate boundary along the Japan trench is almost fully locked in place and is not sliding1,2. But it has been recognized for long time that, in the Japan trench, the ratio of cumulative fault slip of large earthquakes to plate motion (the seismic coupling coefficient) is only about 30%. There has been no explanation of how the remaining 70% of plate motion is accommodated.

Japanese seismologists had noticed this discrepancy, but we had not seriously considered its potentially disastrous implication. In 2001, Ichiro Kawasaki of Kyoto University suggested that a significant portion of the plate motion is accommodated as 'afterslip' — further slippage after an earthquake has taken place — and other aseismic faulting behaviour in the northern part of the Japan Trench3. But the discrepancy was not resolved for the southern part, the main source region of the magnitude-9 earthquake.

Thus the second lesson of the great Tohoku earthquake is that we should not overlook inconsistent data, but instead strive to integrate observational information with different temporal and spatial scales.

Earth science is multidisciplinary. The Japanese seismology community now needs to review all the seismic, geodetic, geomorphological and geological data to find information missing from the current evaluation, and to resolve any inconsistencies.

Prepare for the unexpected

Hiroo Kanamori

California Institute of Technology, Pasadena, California

The 2011 Tohoku earthquake caught most seismologists by surprise because no earthquake of a magnitude greater than 8.5 was known to have occurred in this region. Extensive analyses of seismic, GPS and tsunami data conducted since the earthquake make it clear that unusually large strain and stress release occurred in a relatively narrow zone within 150 kilometres of the Japan trench (see 'Pressure zone'). The amount of strain release is nearly an order of magnitude larger than what we have seen in other mega-thrust earthquakes. The strain must have accumulated in this zone for nearly 1,000 years, with the plate convergence rate of about 9 centimetres per year. Finally, the stress exceeded the local strength of rock and failed, causing the magnitude-9 earthquake.

Despite the extensive GPS network in Japan, this localized large strain build-up had not been detected because the area is 200 kilometres offshore. An important lesson we have learned is that such a large strain can accumulate in the shallow plate boundary (see 'Pressure zone') — it had been considered able to accommodate only 10–20% of the strain released during the 2011 event.

Thus, for monitoring a strain build-up large enough to eventually cause a magnitude-9 event, it is important to accelerate the effort to develop ocean-floor GPS technology, and to promote research to detect unusual seismogenic structures responsible for such large strain build-up (for example, rough plate interfaces).

Even if we understand how such a big earthquake can happen, because of the nature of the process involved we cannot make definitive statements about when it will happen, or how large it could be. It will be a very rare event but, once it happens, it will have grave consequences. So we must try our best to be prepared for the unexpected. Building a robust infrastructure is most important. However, there is a limit to what we can do with the available technology, so we need to seriously consider the acceptable trade-off between benefit and risk.

Enhance ocean-floor observation

Yuji Yagi

University of Tsukuba, Japan

The process by which the Pacific plate is subducting beneath the Japanese archipelago is not smooth. Strain can accumulate where parts of the plate are unable to slip because they are stuck to the overlying continental plate. Major inter-plate earthquakes release the resulting accumulated strain, allowing slip to resume. The 2011 Tohoku earthquake released a huge slip deficit, which was revealed at least in 2004 (ref. 1), or even as far back as 2000 (ref. 4) by the GPS network, and anticipated from a discrepancy between geological and geodetic rates of crustal deformation by Yasutaka Ikeda of the University of Tokyo in 1996 (ref. 5).

The distribution of areas where the plates are stuck (known as asperities) is usually inferred from the analysis of seismic event data. And, on the basis of this 'characteristic earthquake model', the regions likely to experience future earthquakes are estimated by responsible commissions in Japan such as the Headquarters for Earthquake Research Promotion. Most seismologists in Japan assume reasonably that areas beyond the asperities are free from earthquakes for long periods of time because the strain there is released by aseismic slip. We need to confirm this assumption.

Great earthquakes occur not only as a result of the combined effect of simultaneous strain release at several assumed asperities. The accumulation of strain suggested by crustal deformation or GPS observation has not been adequately considered in the government's seismic-hazard estimations in Japan because of the low resolution of the data. This oversight reduces the total amount of assumed slip deficit. This is why most seismologists did not recognize the risk of a huge Tohoku earthquake.

We are not able to obtain accurate information about the patterns of great earthquake occurrences in time and space because of the long period of the seismic cycle. Therefore, we cannot assess the probability of future great earthquakes on the basis of catalogues of recent seismicity alone. Great earthquakes are exceptional events reflecting very long-term deformation processes in a subduction zone. To develop a comprehensive understanding of them we must also consider geodetic, geological and geomorphological information about crustal deformation.

In general, for predicting long-term seismicity, methods must be developed to extract information about elastic and anelastic strain from observational data. Pressingly, ocean-floor observation must be enhanced to allow high-resolution estimates of slip deficit distributions along the plate boundaries.

Warnings work, but must be better

Masumi Yamada

Disaster Prevention Research Institute, Kyoto University

The Japan Meteorological Agency has one of the most advanced systems in the world for providing real-time warnings of tsunamis and earthquake shaking. The earthquake early warning system, which provides information about strong shaking within seconds of a quake, has been in place since 2007 and has provided more than 10 warnings of strong earthquakes — by cellular phone, television, radio and local-community speaker system. But it could be better.

The system detected the earthquake off the Pacific coast of Tohoku and, about 8 seconds after the first primary wave arrived at the closest seismic station, issued a warning to the public in the region close to the epicentre. Twenty seven bullet trains were stopped without derailments in this region. Three minutes later, warnings for very large tsunamis were issued to Iwate, Miyagi and Fukushima prefectures. The damaging waves arrived 15–20 minutes later at the closest shores.

However, the overall performance of the system was not satisfactory, mainly because of the complex character and relatively small amplitude of the beginning of the rupture. The system underestimated ground motion and tsunami heights, so the large population in the greater Tokyo region, where many areas experienced strong and damaging shaking, received no warning (see 'False comfort'). That said, updates did improve as more information became available.


Early warnings for strong shaking were broadcast more than 70 times for aftershocks. The system worked well for these smaller events, but there were some errors in determining event locations because of the complication of simultaneously occurring earthquakes.

The unexpected character of the seismic data at the start of the earthquake fooled the early warning system's algorithms. But the system has the potential to work well for the next great earthquake — such as the widely expected Nankai earthquake in the Kansai region — if technical improvements are made to recognize great earthquakes quickly. The earthquake early warning system in Japan should become a truly effective mitigation tool in a society that has already accepted and learned to expect such information.

Design buildings for greater shakes

Jim Mori

Disaster Prevention Research Institute, Kyoto University

The Tohoku earthquake came as a frightening and disheartening surprise to Japanese seismologists, who thought they could roughly predict the locations and sizes of plate-boundary earthquakes along the subduction zones of Japan. Some 400 years of historic records have proved to cover too short a time period to be a reliable guide to the occurrence of the largest earthquakes, even for this very seismically active region. With hindsight, we must now re-examine the Japanese geological data for events of a similar magnitude.

Usually in seismology, the rates at which future earthquakes are expected to occur have been largely derived from the statistics of repeating events on faults.

However, the 2011 earthquake shows that rarer and much larger earthquakes can also occur in the same fault zone as smaller events. So past statistics are not always sufficient. One way to improve our understanding is with efforts to measure the local accumulation of stress near faults and to estimate the absolute stress and strain levels at which earthquakes happen.

For example, stress can be measured directly within boreholes; frictional strength can be inferred from temperature measurements after large earthquakes; and directions of regional stress fields can be determined from patterns of ground displacements recorded by GPS arrays.

Using such techniques, we may be able to determine how close a fault is to failure, and thus estimate the earthquake risk more directly. The size of an impending event will still be difficult to determine, but maximum sizes might be estimated from the stress measurements. The regional deformation of eastern Japan is now being studied to try to discern what stress and strain the area underwent before the 2011 quake.

The evaluation of strong shaking from great earthquakes is a related issue of great importance — for building safety standards around the world — but a long record of accurate ground motions from past earthquakes is lacking. As measurements have improved over the past few decades, our expectations of the severity of the motions have steadily increased. It is unlikely that we have seen the worst. Greater shaking, especially at long periods of several seconds, is a possibility that must be considered in hazard planning. Such information is needed for the safe design of tall buildings in our modern cities.