We review long-term forecasts of great earthquakes along subduction zones around Japan and discuss the related problems from a paleoseismological point of view. Rich historical data in Japan show the recurrence of great earthquakes along subduction zones, particularly at the Nankai trough, for more than 1, 000 years. On the basis of such historical data and interseismic/coseismic vertical crustal movements, Imamura made a rather vague forecast of great earthquakes along the Nankai trough as early as 1933, which turned out to be successful by the occurrence of the 1944 Tonankai (Mw 8.1) and 1946 Nankai (Mw 8.1) earthquakes. After plate tectonics theory was established, the concept of seismic gap in subduction zones has been thought as a powerful tool for long-term earthquake forecasts. Great interplate earthquakes have been predicted from examinations of not only seismic gaps but also other observations such as seismic quiescence, earthquake recurrence history, current crustal deformation in coastal areas, or seismic crustal movements in geologic records. The 1973 Nemuro-oki earthquake (Mw 7.8) along the Kurile trench was predicted in 1972, although it was slightly smaller than the predicted size. The 1978 Miyagi-oki earthquake (Mw 7.6) along the Japan trench was also predicted in 1977, although the size and place were somewhat different from the prediction. The Tokai earthquake was predicted in 1976, and its occurrence has been considered imminent. Despite that short-term surveillance system has been in operation in the last 20 years, this earthquake has not occurred yet. In 1994, two great/large earthquakes occurred in subduction zones off northeast Japan, but no forecasts had been made on these. The off-Hokkaido event (Mw 8.2), whose aftershock area apparently coincides with that of the 1969 interplate earthquake (Mw 8.2), has been interpreted as an intraplate event within the subducted Pacific slab. The Sanriku-oki earthquake (Mw 7.7) was aninterplate event, but its rupture zone overlapped with a previous interplate event, the 1968 Tokachi-oki earthquake (Mw 8.2). The recurrence history of interplate earthquakes along the Nankai trough has been updated by seismo-archaeologlcal data such as liquefaction evidence at archaeoiogical sites. and it now seems more regular and can be explained by time-predictable model. In the mean time. historical data indicate that the 1605 earthquake was an unusual “tsunami earthquake” and the source process is very different from the other repeated events. This event may have been affected by a preceding large inland earthquake in 1596. Paleoseismological investigation is still very important for long-term forecast of earthquakes. Future research should emphasize to complement recurrence history of interplate earthquakes from historical as well as seismo-archaeological data, to distinguish intraplate (slab) earthquakes from interplate earthquakes in historical catalog, to investigate the co-relation between interplate and inland earthquakes, and to combine historical, geologic and other kinds of data to study earthquake recurrence as demonstrated in the Cascadia subduction zone.
A long-term prediction of earthquakes for the ltoigawa?Shizuoka Tectonic Line Active Fault System was made public in 1996 by the Earthquake Research Committee of the Headquarters of the Earthquake Research Promotion. The prediction is that an earthquake of magnitude about 8(M 71/2-81/2) will occur within a few hundreds years with high probability from the Itoigawa?Shizuoka Tectonic Line Active Fault System including the Gofukuji segment, though the extent of the seismogenic segment can not be specified. This statement was based on the following evaluation on the past activity of the fault system: the last faulting event was about 1200 yrs ago and the surface rupture zone at that time extended over 100 km from Hakuba to Kobuchizawa, and the slip was 6?9 m at Gofukuji and about 6 m at Chino. The recurrence interval of faulting at the Gofukuji segment was estimated at about 1000 years from ages of the last three events, which is consistent with the long-term slip rate, 8 mm/yr, of the segment. The magnitude of the next earthquake was estimated from the rupture length (about 100 km) and the amount of slip (6?9 m) in the last event and the time elapsed since the last event. The probability of the earthquake occurrence was estimated based on the step-model in which an active fault at a point or a segment moves with a characteristic displacement D and a definite recurrence interval R, as shown in relation of R=D/S, where S is the long-term slip rate on the fault. When R is close to time (t) elapsed since the last faulting, we think the probability of the occurrence is high. In this case, R is about 1000 years, and t is about 1200 years. So, the probability of the occurrence is expressed in the statement as high. A preliminary estimation, although it was not included in the statement, shows the probability of occurrence of the next earthquake is about 30% in coming 100 years, assuming the lognormal distribution of variability in recurrence time and standard deviation 0.3 in the renewal model.
The Itoigawa?Shizuoka tectonic line active fault system (ISTL) is one of the longest and the most complex active fault systems on land in Japan with very high activity. The system comprises the northern (55 km long east dipping reverse faults), the middle (60 km long left-lateral strike-slip faults), and the southern (35 km long west-dipping reverse faults) sections. The estimates of the average slip rate range 2 to over 10 m/103 yr in the system. This high slip rate and probable quiescense of the system exceeding 1150 years indicate the possibility of a surface faulting event in the near future. Since historic and instrumental records of seismicity along the ISTL is very poor, geological study on the paleoseismology of the ISTL has an important clue to evaluate the long-term seismic risks of the fault zone. In 1995 and 1996 the Geological Survey of Japan opened six exploratory trenches in the fault system and the results from the three in the northern section are reported in this paper. The Hakuba trench on the Kamishiro fault brought four earthquake events since 6738 BP (dendrocorrected radiocarbon age in calendar year) with the average recurrence interval to be between 1108 and 2430 years. The last event here postdates 1538 BP. The Omachi trench exposed the last event after 6th to 7th century AD and before 12th century at the latest, Only one event after 3rd to 4th century AD was identified in the Ikeda trench. The timing of the last event from each trench is between 500 and 1500 BP, which interval coincides with the timing of the last event in the middle section as well as the 841 AD or 762 AD earthquake reported in historical documents. The dating of the upper age limit of the last event is not precise enough to correlate the event with any of known earthquake. The recurrence interval of the northern section, however, is significantly longer than that of the Gofukuji fault. The difference in the recurrence time from one section to another is concordant with the difference in the apparent slip rate.
The aim of this study is to show the long-term conditional (time-dependent) seismic hazard assessment of intraplate earthquakes in Japan. Two types of hazard assessment maps are illustrated by the combination of statistical method, long historical earthquake records, and the results of trenching surveys: one is the probability map of 0.2 g between 2001 and 2100 A.D., and the other is the peak horizontal ground acceleration (PGA) map of 10% chance including the effect of attenuation of surface geology during the same period. Although the number of trenching results with three or more events are only nine as of 1994 and the error of the data is large, the results still enable us to describe that (1) the compilation of the paleoearthquake results indicate that the recurrence interval of each fault seems to be quasi-periodic than random, and (2) the entire length of fault system does not always rupture together. This is important for seismic hazard assessments because the size and the frequency of earthquakes vary as a function of fault length. Earthquake statistics is then applied to the results of trenching surveys in order to calculate the conditional probabilities of active faults. For each trenching study site on a fault, I calculate the average recurrence interval, Tave. between events and then determine the ratios of the recurrence interval for specific event, T, pairs to Tave. The resulting distribution of T/Tave measurements are the basis to determine Weibull probability density function. The initial analysis suggest that the application of conditional estimates on specific faults is advantageous over the assumption of random process. The resulting maps show better agreement with the current understanding of recurrence cycle of active fault that, (1) low probabilities are obtained for faults that are considered to have ruptured in historical period, and (2) higher probabilities are calculated on faults with long elapsed time or high slip rate. Regarding the PGA map, the 0.5 g or higher are calculated at plains and basins bounded by the active fault with high probability, especially on the Median tectonic line, the Itoigawa-Shizuoka tectonic line, the Inadani fault system, the Kozu-Matsuda fault system, and the Beppu-Haneyama fault system. The maps are preliminary and will be modified by including the future paleoearthquake results.
Recent studies (1987 and later) of seismicity patterns presumably having some power to predict large earthquakes are reviewed. Here ‘large earthquakes’ mean relatively large earthquakes considered as objects of prediction. Seismic gaps (observed in the later half of the recurrence cycle) are useful in limited fault segments. Many examples of seismic quiescence (local or regional) have been reported, some of which were followed by large earthquakes and others were not. Migration patterns of seismic activity are very variable and mostly transient. There are some cases of close correlation between large earthquakes in two regions. The test of the statistical significance of a proposed pattern is often difficult due to uncertainty in evaluating the effect of selection of favorable data set. However, many of the patterns seem to be real and can be used in the long-term probabilistic prediction of large earthquakes. The probability of occurrence of a large earthquake after the appearance of one of the patterns discussed here is not easy to estimate. This is because of (1) many parameters which characterize the pattern and the target earthquake, (2) regional differences in the character of pattern, and (3) relatively few observational data on each pattern in each region.
This article reviews observational and theoretical studies on seismic gap and seismic quiescence from the view-point of earthquake forecast. “Seismic gap of the third kind” by Y. Ishikawa is also briefly discussed. Seismic gap, which is recognized as an unruptured segment of plate boundary and large-scale geological structure in the present seismic cycle, is a strong tool for identifying the zone of high seismic potential in the near future. However, the concept of characteristic earthquake is not always valid. We need reliable information on the duration of seismic cycle, slip distribution of past earthquakes, and seismic coupling for effectively applying the seismic gap hypothesis to earthquake forecast. Seismic quiescence has been reported prior to major seismic events of a wide magnitude range, from great earthquakes to rock fractures in the laboratory, but the physical mechanism is not yet established, Among the several hypotheses so far proposed, the stress relaxation model based on the laboratory-driven friction law is of particular interest since it theoretically predicts the formation of seismic quiescence as a natural consequence of the physical process of earthquake preparation. For checking the validity of the model, we need detailed studies on the time-space development of quiescence, and focal mechanism change in the surrounding area. In spite of some limitations described above, past studies demonstrate that a combined use of the seismic gap and seismic quiescence is expected to provide a most useful tool to forecast the occurrence of a large earthquake for the time range of a year to several decades.
Temporal changes of the hypocentral distances of earthquakes around the major shocks in and around Japan were studied based on the JMA (Japan Meteorological Agency) hypocenter catalog to examine a hypothesis that the area of the precursory seismic gap expands with time until a mainshock occurs at the center. Most of the inland large earthquakes occur in the interior of long-term seismic gaps. The most noticeable case is that the gap expands after the occurrence of a small-scale seismic activity around the center of the gap. The radius of the gap is approximately represented by an exponential function of time with the time-constant being 3?4 years for major inland large earthquakes in Japan. The conditional probability that the above case occurs prior to a large earthquake is 26% for events of the inland earthquakes with magnitude 5.8 or more. The probability is raised up to 50% for the events with magnitude 7.0 or more. The gap expands 4?5 times as large as the radius of the source region of the main shock. For the regions close to a plate boundary such as the Japan trench east off north Honshu, the time-constant is as small as 1?2 years and the gap area is confined to the same range of the mainshock source region. The above time-constant is inversely proportional to the expansion rate of seismicity gap and is considered to be closely related to stress relaxation process and/or hardening of the medium. Although the expansion phenomenon of a seismicity gap is not always followed by a large earthquake, it may be an important factor for estimating the possibility of a large earthquake.
The anomaly in focus is relative quiescence which is defined as a significant decrease of earthquake activity compared with the predicted occurrence rate by the estimated statistical model for the standard seismicity of the region: we use a point-process model called Epidemic Type Aftershock-Sequences (ETAS) model. Regardless of the seismicity level, the relative quiescence can take place. Size of such quiescence is seen in cumulative and M-T diagrams of transformed occurrence times by using the estimated ETAS model. By this method, we recognized significant, relatively quiet stages in shallow seismicity over M 5 class in a very wide area preceding all studied great earthquakes of M 8 class in and around Japan and also those of M 9 class in the world. However, we saw no relative quiescence for about 20 years up to 1990 in the wide areas which include the Tokai and Boso gaps [OGATA (1992)]. On the other hand, a few authors recently showed a very significant quiescence for the last 20 years since early or mid 1970's in the seismicity of certain areas in or around Tokai region. However, it turns out that these quiescence are seeming ones owing to magnitude shifts which took place during 1975?76. The magnitudes below MJ5.0 are substantially underestimated after the period. This shift is found and estimated by a statistical comparison of magnitudes between the JMA and USGS catalogs. Nevertheless, the recent seismicity with level of MJ≥5.5 in a very wide region including central and western Japan shows a significant relative quiescence since 1992, which may be related to the 1995 Kobe Earthquake of MJ7.2 but seems to be further extended to late 1996 when many M 6 class events successively occurred about the boundaries of the quiet region within a half year span.
This paper first reviews recent studies on constitutive formulations for shear failure. Noting that earthquake rupture is an general a mixed process between what is called frictional slip instability and fracture of initially intact rock in the fault zone, the paper then discusses how the constitutive law for earthquake rupture should be formulated to meet the physical principles and constraints to be imposed on the law. This leads to the conclusion that the constitutive law should be formulated as a slip-dependent law, which includes a scaling parameter Dc (critical slip displacement). Scale dependent physical quantities inherent in shear rupture, such as the breakdown zone size and the nucleation zone size, their duration, the slip acceleration, and the shear fracture energy, are consistently scaled by Dc, which in turn is prescribed by a characteristic scale representing the geometric irregularity of the fault zone. A physical model for earthquake generation is presented, which reconciles the observation that the mainshock seismic moment is scaled by the critical size of the nucleation zone. The constitutive law parameters depend on ambient conditions in the seismogenic layer, and therefore it is critical to evaluate the effects of the environmental factors on the constitutive law parameters in quantitative terms, in order that numerical simulation of a model earthquake could be useful for earthquake prediction in the real world.
Earthquakes are complex phenomena. The notion of self-organized criticality (SOC) is recently used to explain various kinds of seismological relations expressed as power-laws, such as the size-distribution, the fractal spatio-temporal distribution and the decay of aftershocks. It is often claimed that earthquakes are not predictable if they are critical phenomena It is true that they are not predictable if the earth is at exact criticality. The notion of SOC. however, is not static but is dynamic. The crust gently approaches criticality and breaks down to generate a large earthquake. There are a variety of SOC models of earthquakes. In all of them, most of small and intermediate earthquakes occur while the system is at subcritical state approaching the exact criticality. Large earthquakes mostly occur near the criticality or at the supercritical state. Therefore, large earthquakes differ from small and intermediate ones even in the statistically scale-free SOC models. Large earthquakes in nature also are expected to be unusual events occurring at critical or supercritical states at which physical properties are extemely abnormal. Different SOC models exhibit different precursor phenomena before large events, as observed precursors differ from event to event. There cannot be universal methods to predict every large earthquake, but individual large earthquakes can be expected to have some precursors. It may be careless to reject observed precursors on a reason that they were not always observed before large earthquakes.
A model called strength distribution model is proposed for elucidating the fracturing process of intact rock specimens under tri-axial loading of compression. This model provides a way to estimate the fracture stress from the relationship between crack density and applied stress for the specimens. Making use of the model, the fracture stress is estimated from the observed relationship in the range of applied stress from the onset of loading to the arbitrary upper bound. The estimated stress is found to be in agreement with the observed one with an error smaller than 1%, whenever the upper bound is taken to be more than 95% of the fracture stress. This model is also applied to the seismic activities prior to the 1989 Loma Prieta earthquake and to the 1996 Adak earthquake. The occurrence time and the magnitude respectively are estimated to be almost equal to the observed ones for both the earthquakes, Although more studies are needed to conclude, these results imply that earthquakes occur in principle following the same law as fracturing of intact rocks. The results of the fracture stress estimation for intact rocks may suggest that the long-term prediction of earthquakes is possible several years before the target earthquake with an error of about one year in the best case, provided that the recurrence time of the earthquake is about a hundred years.
We analyze the dynamics of two interactive parallel faults on the basis of a discrete dynamical model. This fault model can be regarded as an extension of the Burrige-Knopoff model. Our main concert is about the temporal variation of characteristic events on the two interactive faults. The tendency is observed that the characteristic events continue to occur only on one of the faults for a period of time; the occurence is quasiperiodic in this period as generally found for the Burridge-Knopoff model. However, the activity is suddenly transferred to the other fault at some time. The period of characteristic even occurrnece tends to alternate on the two faults. This length of this period is generally lager when the interaction are weak between the two faults. This suggests that the recurrence of large events on neighboring faults is highly complex. It is also indicate from our analysis that long apparently quasiperiodic recurrence of large events on a faults may cease at some time if weak fault interaction exist.
Information on spatial and temporal distribution of stress state and fault strength within seismogenic zones is essential for improving the accuracy of long-term forecasts of large earthquakes. We would be able to specify the area with a high probability of earthquake occurrence, if we know distribution of highly stressed areas in comparison with their strengths. Although at present we don't have a method to measure stress and strength within the seismogenic zone and it is extremely difficult to establish the method, we need to make every effort to extract information on stress and strength and that on their fluctuations in time and space through observations and surveys.
It is useful to express the state of crustal stress by the ratio (μm) of maximum shear stress to mean stress because the effects of depth on stress are compensated. Using the changing rate of μm calculated from repeated measurements of stress and present stress state in the upper crust, and assuming that shear fracture limit μc is from 0.6 to 0.7 by the Coulomb criterion and Byerlee's law, middle term (10-odd years to 1 year) prediction of earthquake might be possible. In the case of the 1995 Hyogo-ken Nanbu Earthquake, μm increased since around 1983 at both the Hiraki and Hoden stress measuring sites, and was estimated to have reached to 0.6 at the end of 1994. Measurements after the earthquake at both sites showed a decrease to the previous level of μm=0.2. Two months before the earthquake, crustal strain observation at the Rokko-Takao station indicated a decrease of the stress normal to the strike of the Rokko faults system, and at the same time the amount of seepage of under-groundwater increased in the Rokko-Takao observational tunnel. This also suggests an increase of hydraulic pore pressure in the fractured zone of the faults at depth.
Physical processes of earthquake generation may be divided into three different stages, such as tectonic loading, quasi-static nucleation, and subsequent dynamic rupture propagation. The basic equations governing the earthquake generation process are; the equation of motion in elastodynamics that relates slip motion on a fault surface with deformation of the surrounding elastic media, the fault constitutive law that prescribes the relation between shear stress and fault slip (and/or slip velocity) on the fault surface during earthquake rupture, and the loading function that gives the increase rate of eacternal shear stress induced by relative plate motion. Recent development in physics of earthquake generation enables us to simulate the entire process of earthquake generation by solving these nonlinear coupled equations. For long-term prediction of earthquake occurrence the detection of crustal movements associated with tectonic loading is important. For short-term prediction the detection of precursory phenomena associated with rupture nucleation is important. In either case it is necessary to establish an interactive forecast system based on theoretical simulation and continuous monitoring of earthquake generation processes.
We synthesize temporal variation in Coulomb Failure Functions (CFF) for the mechanism similar to the 1995 Hyogo-ken Nanbu earthquake taking both interseismic stress loading and coseismic changes due to large events into account. A block-fault model derived from the inversion of triangulation and trilateration data during 100 years by HASHIMOTO and JACKSON (1993) is adopted as the model for interseismic loading. Temporal changes in CFF is calculated as a sum of secular changes for back slip of block boundary faults in their model during arbitrary period since 1855 and coseismic changes due to large events which occurred during the corresponding period with simple pure elastic dislocation model. Among the several factors which affect CFF change in the source region of the 1995 Hyogo-ken Nanbu earthquake, increasing back slip of faults along the Nankai trough, which may represent strengthening of coupling, decreases CFF for right-lateral strike-slip mechanism on NESW trending vertical faults by 0.2 MPa/100yr at (34.6°N, 135.04°E and 10 km deep). Back slip along the Arima-Takatsuki Tectonic Line and Rokko fault system increase CFF most (1.6 MPa/100yr), though the calculated point is closely located to the edge of modeled block-boundary fault. These calculations seem contradictory to the hypothesis that strengthening of coupling between the subducting Philippine Sea plate and the Japanese islands induces the activity of large inland events. Loading of locked part due to stable sliding of deeper part of block boundary faults may play a main role for the accumulation of stress. Among large events, the 1946 Nankaido earthquake may have caused eminently large increase in CFF (0.1 MPa) near the epicenter of the Hyogo-ken Nanbu earthquake. The 1944 Tanankai earthquake may have increased CFF by about 0.01 MPa, but the 1927 Tango earthquake decreased it by 0.02 MPa. Other events may not have larger effect than 0.01 MPa. Summing up these contributions, our results suggest that the CFF for the Hyogo-ken Nanbu earthquake increased larger than 2 MPa since 1855.
An earthquake of M 4.3 occurred in the central area of Shizuoka prefecture on Oct. 5, 1996. The event, including the main shock and its aftershocks, presented strange characteristics in the hypocenter distribution pattern, and also in the focal mechanism pattern. The hypocenters occupy a particular position where seismic activity was rarely observed, the place supposed to lie in the zone of the plate boundary between the subducting Philippine Sea plate and the overriding one. It is considered that both plates are strongly coupled and that their relative motion is locked at that point, and consequently such locking may be the mechanical source of a future Tokai earthquake. The focal mechanism of the main shock is low angle thrust type in accordance with the assumption of the locking. By contrast, those of the aftershocks presenting normal type faultings imply that there is no longer any strong locking. This difference in the focal mechanism between the main shock and aftershocks leads to the supposition that a weakening process of locking progresses at that site. Similar phenomena were found in an investigation of focal mechanisms in a wider area ranging over the focal zone inferred for a future Tokai earthquake. Since the beginning of the 1990s, normal type mechanisms appear to prevail in the shallower side of the focal zone. In this case, there is a concern about the possibility of pre-slipping, that is, a gradual slip starting from the deeper side of the focal zone of a forthcoming earthquake.
Most of large main shocks have preliminary ruptures, and their duration times are proportion to the earthquake magnitude. This relationship indicates the initial rupture process determines the final earthquake size of main shock. The dynamic crack-crack interaction is considered for the growth process of the earthquake. In the region with high crack density, any crack has a lot of chances of rupture initiation, but almost of ruptures are decelerated and stopped by the strong interaction. In this case, each rupture remains a small event. On the contrary, in the case of low crack density, a crack grows up quickly and becomes a large earthquake, because of weak interaction. We assume the crack density is controlled by the fluid which is distributed widely in the lower crust. The upper extent of fluid body can be detected by the reflected seismic waves. About three months before the 1984 western Nagano earthquake, the percentage of the events with clear reflected waves increased and the top of the fluid came up shallow. In this period, the seismic activity was very low, namely the source region was in the stale of seismic quiescence. In most case, the rupture of large main shock initiates from the sharp change portion of the lower boundary of the seismogenic zone. Besides, the depth variation in lower boundary is proportion to that of the upper extent of reflector. Therefore, we can find the initiation region of large main shock by surveying the sharp depth-change area of reflectors, which lie several km beneath the cutoff depth of seismicity. We can also catch the initiation period of large main shock by monitoring the seismic quiescence and reflection efficiency of seismic waves.
The role of the minimum horizontal compressional stress (σhmin) in the deformation of the Japanese Islands and the occurence of intraplate earthquakes were discussed. The first is the effect of decreasing σhmin, which enlarges the Coulomb failure stress. The second is the effect of an absolute value of σhmin. Large σhmin can promote a plastic flow in the upper crust resulting from a subcritical crack growth of microfractures when the maxmum horizontal compressional stress (σhmin) is small, because large σhmin and small shear stress (τ) makes cracks to be shutten and then more cracks are filled with pore water when there is less water in the upper crust. It is thought that the northeast Japanese Island (outer arc of the Tohoku district) is situated under such a stress condition. A larger part of the deformation there is probably shared by a plastic flow.
For long-term prediction of large earthquakes at plate boundaries, it is important to develop an earthquake cycle model which can numerically simulate the processes of crustal deformation and stress accumulation at and around plate boundaries. We developed a general earthquake cycle model applicable to transform fault zones, subduction zones, and collision zones on the basis of elastic dislocation theory. In this model we regard the periodic occurrence of interplale earthquakes as a perturbation of the hypothetical steady slip motion proceeding on the whole plate boundary. Then the crustal deformation associated with the periodic occurrence of interplate earthquakes is given by the superposition of viscoelastic responses to the steady slip on the whole plate boundary, the steady back slip on the seismic zone, and the sequence of periodic seismic slips. The important points in this earthquake cycle model are as follows. Interaction between adjacent plates can be naturally represented by the increase of discontinuity in tangential displacement across the plate boundary. Viscoelastic stress relaxation of the asthenosphere strongly affects the crustal deformation during interseismic period. Since the stress relaxation time of the lithosphere-asthenosphere system is several hundreds years, we must consider the viscoelastic effects due to a series of large events occurred for the past several hundreds years to estimate crustal deformation correctly. Crustal deformation and stress accumulation due to steady slip on the whole plate boundary can not be neglected.
We had the 1989, 1992 and 1994 Sanriku-Oki earthquakes in the region from 39. 0°N to 40. 6°N along the Japan trench. Summarizing the aseismic nature of the earthquakes, we propose two new concepts“seismo-geodetic coupling“and”a gap of the moment release”. The sum of their seismic moments is (4. 8-6. 6) ×1020Nm. These events were associated with aseismic ultra-slow fautings of source times of a day to a year following the respective main seismic shocks. The ultra-slow faultings respectively released interplate momens, 1. 4, (1-4), 4. 2×1020Nm, equivalent to those of Mw7. 3 to 7. 7 seismic events. The sum of the moments released by both seismic and aseismic faultings amounts to (11-16) ×1020Nm. The interplate moment is accumulated as (19-23) ×1020Nm for the area of 180 km (39. 0°N to 40. 6°N) 150 km (width of high seismicity region between the Sanriku coast and the Japan trench), with the relative plate motion of 8 cm/yr and the interval of 21-27 years from the 1968 Tokachi-Oki earthquake of Mw8. 2 to the 1989-1994 ultra-slow earthquakes. The ratio of (11-16) × l020Nm to (19-23) ×1020Nm leads to the rough estimate of “seismo-geodetic coupling coefficient”of 50-85%, where the ordinary seismic coupling coefficient has been estimated to be 20-30% between the Okhotsk and subducting Pacific plates. Based on the spatiotemporal distribution of the interplate moments including the ultra-slow earthquakes, three gaps of the moment release are suggested for the Sanriku-Oki region. This would serve as a basis for the mid-term/long-term prediction of future interplate earthquakes in the Sanriku-Oki region.