Non-technical summary: The Ms=7.9 1923 Kanto earthquake was one of the most destructive events of the 20th century, causing the deaths of over 140,000 people and destroying Yokohama and large parts of Tokyo. Today, the Tokyo metropolitan area houses more than 30 million people. It accounts for a third of Japan's economy and dominates politics, trade, finance, arts and communication. It is not difficult to imagine that a similar earthquake today would not only cause a human tragedy of immeasurable proportions, but would also have catastrophic effects on the Japanese and world economies.

This study presents a simple model for the mechanism of the 1923 Kanto earthquake. To assess the seismic hazard of the Tokyo Bay region, such a model plays a crucial role, since it has important implications for the derivation of repeat times of similar earthquakes that may be characteristic for this area. A unique aspect of our analysis is the use of an unprecedented large set of geodetic measurements of coseismic deformation, consisting of displacements from leveling and angle changes from triangulation surveys between 1883 and 1927. Our final model consists of two adjacent low-angle planes accommodating a combination of thrust and right-lateral strike-slip motion of 7 m (see figure). The geometry of the planes agrees with results from a recent seismic reflection study.

The earthquake was located in the Sagami trough, where the Philippine Sea plate subducts under Honshu at a rate of 30 mm/yr (measured with GPS). The time interval necessary to accumulate the 7 m of slip at this plate motion rate agrees with the actual time interval of 220 years between 1923 and the 1703 Ms=8.1 earthquake, the previous event that ruptured the same plane. However, uplift records of marine coastal terraces in Sagami Bay document 7,500 years of earthquake activity and estimate average repeat times of 400 years for events with similar vertical displacement profiles as the 1923 earthquake. This means that on average a 1923-type event occurs every 400 years and, with a slip of 7 m, only releases about 50% of the total relative plate motion that accumulated in those 400 years, i.e. about 15 m of convergence. In other words, the average slip deficit per recurrence interval is about 50% of the relative plate convergence.

Compared to the highly oblique angle of plate convergence the coseismic slip has a more orthogonal orientation to the strike of the plate boundary. This observation, known as plate motion partitioning, suggests that perhaps other faults or tectonic processes also play a role in releasing the accumulated plate motion.

The findings of plate motion partitioning and slip deficit lead us to suggest that, instead of a simple recurrence model with characteristic earthquakes, additional mechanisms are necessary to describe the accommodation of deformation in the Kanto region. So far, obvious candidates for these alternative mechanisms have not been discovered.

Our two-plane uniform slip source model with isodepth contours for both planes and isodepth contours of the Philippine Sea Plate (PH) after Noguchi [2000].