The Pacific plate converges with northeastern Eurasia at a rate of 8-9 m per century along the Kamchatka, Kuril and Japan trenches. Along the southern Kuril trench, which faces the Japanese island of Hokkaido, this fast subduction has recurrently generated earthquakes with magnitudes of up to approximately 8 over the past two centuries. These historical events, on rupture segments 100-200 km long, have been considered characteristic of Hokkaido's plate-boundary earthquakes. But here we use deposits of prehistoric tsunamis to infer the infrequent occurrence of larger earthquakes generated from longer ruptures. Many of these tsunami deposits form sheets of sand that extend kilometres inland from the deposits of historical tsunamis. Stratigraphic series of extensive sand sheets, intercalated with dated volcanic-ash layers, show that such unusually large tsunamis occurred about every 500 years on average over the past 2,000-7,000 years, most recently approximately 350 years ago. Numerical simulations of these tsunamis are best explained by earthquakes that individually rupture multiple segments along the southern Kuril trench. We infer that such multi-segment earthquakes persistently recur among a larger number of single-segment events.

[1] We inverted 13 tsunami waveforms recorded in Japan to estimate the slip distribution of the 1952 Tokachi-Oki earthquake (M 8.1), which occurred southeast off Hokkaido along the southern Kuril subduction zone. The previously estimated source area determined from tsunami travel times [Hatori, 1973] did not coincide with the observed aftershock distribution. Our results show that a large amount of slip occurred in the aftershock area east of Hatori's tsunami source area, suggesting that a portion of the interplate thrust near the trench was ruptured by the main shock. We also found more than 5 m of slip along the deeper part of the seismogenic interface, just below the central part of Hatori's tsunami source area. This region, which also has the largest stress drop during the main shock, had few aftershocks. Large tsunami heights on the eastern Hokkaido coast are better explained by the heterogeneous slip model than previous uniform-slip fault models. The total seismic moment is estimated to be 1.87 Â 10 21 N m, giving a moment magnitude of M w = 8.1. The revised tsunami source area is estimated to be 25.2 Â 10 3 km 2 , $3 times larger than the previous tsunami source area. Out of four large earthquakes with M $ 7 that subsequently occurred in and around the rupture area of the 1952 event, three were at the edges of regions with relatively small amount of slip. We also found that a subducted seamount near the edge of the rupture area possibly impeded slip along the plate interface.

Geologic evidence has shown that unusual tsunami deposits are traced as high as 18 m above the current sea level or as far as 1-4 km inland from the shoreline on the Pacific coast of eastern Hokkaido, and that such unusual tsunamis have recurred at about 500 year interval with the most recent event in the 17th century. We computed coastal tsunami heights along the Hokkaido and Sanriku coasts and inundation at five coastal marshes in Hokkaido where the tsunami deposits were mapped. Three types of faults were tested: giant fault, tsunami earthquake and interplate earthquake models. The giant fault model, with the largest seismic moment, yields the lowest tsunami heights and smaller inundation than the distribution of tsunami deposits in Hokkaido, while the tsunami heights are largest in Sanriku. The tsunami earthquake model yields little inundation in Hokkaido and the smallest heights in Sanriku. The interplate earthquake model produces the largest tsunami heights and inundation in Hokkaido, reproducing the distribution of tsunami deposits on the Nemuro coast. The multi-segment interplate earthquake with variable slip (10 m on Tokachi and 5 m on Nemuro segment) can reproduce the distribution of tsunami deposits on the Tokachi coast as well, and considered as the best source model for the 17th century tsunami, although the Sanriku tsunami heights are more than 3 m, exceeding an inferred detection threshold of historical documents. The seismic moment is estimated as 8 × 10 21 N m (M w 8.5). Comparison with the recent 2003 Tokachi-oki earthquake indicates that the 17th century tsunami source was longer and located further offshore at shallower depth.

The 1896 Sanriku earthquake was a typical 'tsunami earthquake' which caused large tsunami despite its weak ground shaking. It occurred along the Japan Trench in the northern tsunami source area of the 2011 Tohoku earthquake where a delayed tsunami generation has been proposed. Hence the relation between the 1896 and 2011 tsunami sources is an important scientific as well as societal issue. The tsunami heights along the northern and central Sanriku coasts from both earthquakes were similar, but the tsunami waveforms at regional distances in Japan were much larger in 2011. Computed tsunamis from the northeastern part of the 2011 tsunami source model roughly reproduced the 1896 tsunami heights on the Sanriku coast, but were much larger than the recorded tsunami waveforms. Both the Sanriku tsunami heights and the waveforms were reproduced by a 200-km × 50-km fault with an average slip of 8 m, with the large (20 m) slip on a 100-km × 25-km asperity. The moment magnitude M w of this model is 8.1. During the 2011 Tohoku earthquake, slip on the 1896 asperity (at a depth of 3.5-7 km) was 3-14 m, while the shallower part (depth 0-3.5 km) slipped 20-36 m. Thus the large slips on the plate interface during the 1896 and 2011 earthquakes were complementary.

Previous studies indicate that the source area of the 2003 Tokachi-oki earthquake (M 8.0) was smaller than the comparable 1952 Tokachi-oki earthquake (M 8.2) source. We reinverted the 1952 tsunami waveforms, by adopting higher-resolution tsunami simulation method, and estimating and correcting for clock errors of tide gauges from comparison of the 1952 and 2003 tsunami waveforms. The estimated slip distribution indicates that the 1952 tsunami source area was indeed larger than the 2003 source. The distributions of measured and computed coastal tsunami heights also support this conclusion.