Obtaining a sharp image of magma supply path through dense geophysical observations is important for forecasting time and magnitude of hazardous future eruptions. Here we reveal a clear magma plumbing system using dense seismic and geodetic networks around Mt. Asama, central Japan. Magma intrusions occurred several times beneath the western flank of Mt. Asama, forming a WNW-ESE directed zone with 1 km below sea level. The eastern end of this zone connects a narrow vertical pathway extending right under the summit crater, which erupted in 2004. Monitoring magmatic activity with a well-designed observational network is vital to mitigate future volcano hazards.
 Understanding a magma feeding system of a volcano is fundamental to assess a potential volcanic hazard. This goal can be achieved by monitoring magma propagation with well-designed networks of geophysical instruments. Some volcanoes generate various geophysical signals before eruption. For example, long-period seismic signals are often observed at shallow depths [e.g., Chouet, 2003], and earthquakes with higher frequency contents are also observed in volcanic areas [e.g., Rubin et al., 1998]. Also, a volcano deforms before an eruption due to magma injection [e.g., Aoki et al., 1999]. On the other hand, magma plumbing systems beneath some volcanoes generate only weak signals which can be hidden by the signals associated with the migration of fluid in the shallow hydrothermal system [e.g., Battaglia et al., 2006]. In these cases, it is difficult to identify a magma pathway using seismic and geodetic measurements. There have been a few studies to try to delineate the magma system from geophysical observations [e.g., Uhira et al., 2005]. Here we report a sharp image of magma supply path beneath Mt. Asama volcano, central Japan, by integrating precisely relocated hypocenters and ground deformation data obtained from dense geophysical networks. In a way, we were lucky to be able to provide the sharp picture due to the active seismicity and the obvious crustal deformation.
 Mt. Asama, which is one of the most active volcanoes in Japan, is an andesitic volcano located in the center of the country. The summit elevation is 2560 m above sea level, and the size of the active summit crater is 450 m in diameter and 150m in depth. To the west of Mt. Asama, there is a row of older Quaternary volcanoes collectively known as Eboshi Volcanoes. The volcanism near Mt. Asama appears to have progressed eastward, with Asama volcano as its eastern end and the youngest member of the row [Aramaki, 1963].
 At 11:02 (GMT) on 1 September 2004, a moderate-sized eruption occurred for the first time in the last 21 years. From 14–18 September, a continuous stromblian explosion emitted volcanic ash that reached as far as the Tokyo metropolitan area about 130 km away. The volcanic activity seemed to have subsided thereafter, except for moderate-sized eruptions occurring on 23 September, 29 September, and 10 October, and some small-scale eruptions afterward which became smaller and smaller with time. The last moderate-sized eruption occurred on 14 November, and since then no eruption has occurred up to the end of 2005.
 The Earthquake Research Institute, University of Tokyo (ERI) and the Japan Meteorological Agency (JMA) operate seismic networks around Mt. Asama. These two institutions exchange data for the purposes of scientific research and volcanic activity monitoring. Figure 1 shows the spatial distribution of the seismic networks. Two summit stations (KAH2 and KAC2) were broken by the first eruption. The stations coded with four characters (e.g., ASMA), except KAC2 and KAH2, are operated by JMA.
3. Hypocenter Relocations
 Hypocenters of more than five-hundred volcanic earthquakes occurring from 1 January 2004 to 19 October 2005, were determined in the routine data processing by the Asama volcano observatory (AVO), ERI. The routine hypocenters (see auxiliary material) spread vaguely beneath Mt. Asama, and it is difficult to infer a magma supply path from the routine result. We used the double-difference algorithm (DD method) and hypoDD (software) [Waldhauser and Ellsworth, 2000] to obtain the precise relative distribution of these hypocenters. To regularize the ill-conditioned systems of inversion problem, the DD method selects events that are well linked to other events. Therefore, several dozen earthquakes that occurred beneath the eastern flank of Mt. Asama were excluded from the relocated list because these events were isolated from each other. The DD method takes a lower limit of travel-time assortments, so it essentially excludes events with poorly picked data. In this paper, the lower limit of assortments was taken to be eight. Nearly half of the events were relocated, and the main characteristic of the routine hypocenter distribution was included in the relocated one. Using the DD method, Yamamoto et al.  relocated the hypocenters of the earthquake swarm that started at 6:10 (GMT) on 31 August, lasting until just before the first eruption. The stars and solid circles in Figure 2 represent the relocated hypocenters and the relocated swarm, respectively; those are independently determined and plotted on the same figure. The relocated distribution reveals a sharp image of seismicity composed of two groups. One group (Group-I) forms a WNW-ESE directed zone at a depth range between 1 km and 1.5 km below sea level. The eastern end of this seismic zone lies beneath the summit crater and extends westward horizontally over 2 km in length. The other group (Group-II) forms a narrow vertical seismic zone extending from the eastern edge of Group-I to just under the summit crater. The data of the two summit stations, KAC2 and KAH2, were not available after the first eruption. To examine whether the missing summit stations affect the hypocenter distribution, we relocated all these events under the condition that the summit data were excluded from the catalog of travel-time data, and confirmed that the effect of the missing stations was negligibly small. Note that the hypocenters during September 2004 are missing from the relocated results because an extremely high and continuous seismicity associated with the summit eruptions prevented us from obtaining precise hypocenters.
 The hypocenter distribution dramatically changed before and after the eruption on 1 September 2004 (Figure 3). Before the first eruption, almost all events occurred just beneath the summit crater with a depth shallower than 1 km above sea level, and the activity in the deeper part of Mt. Asama was relatively quiet. The hypocenter distribution during this period corresponds to the top 1km portion of Group-II. The hypocenter distribution after the eruption was composed of two groups: Group-I, and the deeper portion of Group-II. Group-I has been activated since the end of October 2004 and almost all of the events in this group are classified into an A-type earthquake: the nature of seismograms is similar to those of the shallow tectonic earthquakes. The most of the events in Group-II are B-type earthquakes. Although they are characterized by emergent onsets, we could pick P-wave onsets at stations near the summit.
4. A Dike Model
 In addition to the seismic data, we also used continuous GPS data. We modelled the ground deformation field between June 2004 to March 2005 by inverting for length, width, depth, dip angle, strike direction, location, and amount of opening of an intruded rectangular dike in an elastic, homogeneous, and isotropic medium [Okada, 1985; Cervelli et al., 2001]. A pressure source model and a fault dislocation model cannot explain the pattern of the ground deformation. The results show that the observations are well explained by a dike intrusion to the western flank of Mt. Asama. The intruded volume is 6.8 × 106 m3, which is about three times larger than the 2 × 106 m3 of magma emitted during the eruption [Nakada et al., 2005]. Even though the total deformations are not large enough to well constrain every parameter, the horizontal location of the dike and the total volume of the intruded dike are constrained relatively well; the estimated standard deviations are ±0.7 km, and ±1.5 × 106 m3, respectively. The eastern part of the dike overlaps with Group-I of the relocated hypocenter distribution (Figure 4). The depth at the top of the dike, whose estimated standard deviation is ±1.3 km, coincides with the depth range of Group-I. The distribution of dike-induced seismicity reflects the distribution of ambient stresses that are near to failure, thus the seismicity might be much more limited in extent than the dike that produced it [Rubin et al., 1998; Rubin and Gillard, 1998], being concentrated near the dike perimeter. The eastern end of Group-I is connected with the narrow vertical seismic zone, Group-II, extending from 1 km below sea level to just under the eruptive summit crater. The swarm activity implies the opening process of the blocked shallowest part of the vent. The relocated hypocenter distribution, in which one part overlaps with the dike and the other part extends vertically to the summit crater, represents the magma supply path beneath Mt. Asama.
5. Discussion and Conclusions
Figure 5 compares temporal variations in the monthly number of A-type earthquakes and that of all volcanic earthquakes with the changes in GPS baseline length between 950221 and 950268 from 1996 to 2004. Because Eboshi-Asama volcanic row lies between these two GPS stations, the baseline extensions between these two GPS sites clearly indicate magma intrusions beneath the western flank of Mt. Asama. The hypocenter distribution before 2004 determined in the routine data processing by AVO represents that the A-type earthquakes occurred under the western side of Mt. Asama and the other volcanic earthquakes occurred beneath the summit of Mt. Asama, although their precision was not as good as that after 2004 due to lack of dense seismic network. The estimated dike location before 2004 was also beneath Eboshi-Asama volcanic raw [Murakami, 2005]. It seems reliable that the overall trend of hypocentral distribution and the dike location had not changed during the last decade. Before the eruption, we recognized three stages of increase in number of A-type earthquakes: the latter half of 1996, from October 2000 to April 2001, and from May 2002 to August 2002. All the three GPS baseline extensions from 1996 to 2003 were synchronized with these seismic activations, suggesting that the A-type earthquakes were associated with the intrusion of magma beneath the western flank of Mt. Asama.
 On the other hand, the GPS baseline contractions were also observed three times: from September 1997 to March 2000, from July 2001 to February 2002, and from March 2003 to April 2004. The baseline contractions indicate migrations of intrusive magma from under the western flank of Mt. Asama to other places. Although the exact direction of the magma migration is unknown because the change in GPS baseline length between 950221 and 950264 is only sensitive to the inflation and deflation under the western flank of Mt. Asama, it would be one of the possibilities of two directions, the direction from which magma is injected into the dike, and the direction to the vent. The activity of the volcanic earthquakes was low during the first contraction period except for the last several months of this period. On the contrary, the activity of B-type earthquakes kept up in high level during the latter two contracting periods. The maximum temperature of the crater had exceeded 200°C from fall in 2002 [Japan Meteorological Agency (JMA), 2005], representing that the shallow part of vent had kept on high-temperature state from the middle of 2002. These observations suggest that there were certain essential differences between the first contraction stage and the last two contraction stages.
 A sudden extension of GPS baseline length between 950221 and ASM4 was detected between 21–22 July 2004, suggesting a nearly vertical magma intrusion into the same dike beneath the western flank of Mt. Asama [Murakami, 2005; Aoki et al., 2005]. Volcanic glows had begun to be observed since the last ten days of July 2004 and the maximum temperature at the bottom of the crater exceeded 500°C after the sudden extension of the baseline length [JMA, 2005]. These surface phenomena indicated that the temperature rise in the shallow part of the vent succeeded the magma intrusion.
 Integrating the observational facts as discussed above, the two baseline contractions after 2000 appear to be caused by the migration of the intrusive magma from the dike beneath the western flank to the vent of Mt. Asama. The gradual supply of magma into the vent had induced a gradual activation of volcanic earthquakes from the middle of 2001 to the first eruption on 1 September 2004. The relocated hypocenter distribution in Figure 3 shows that almost all earthquakes, which occurred from January 2004 to the first eruption, lie in the shallower part of the vent with a depth shallower than 1 km above sea level; during this period, the seismicity was relatively high from May to June 2004. These facts suggest that magma had ascended about 1 km above sea level gradually by June 2004 at least and the ascending magma may have increased the internal pressure in the top portion of the vent causing an enhanced seismic activity at the shallower part of the vent. Before the eruption on September 1, magma migration from the deep chamber to the shallower portion of the vent was relatively slow probably due to the sealing by a cap at the top of the vent. After the eruption, the removal of the cap may have made the magma migration much easier than before and caused the high A-type earthquake activity. This speculation is supported by a fact that partially molten country rocks (rhyolite tuff) are found among the 2004 eruption products [Nakada et al., 2005].
 The dense seismic and geodetic networks in and around Mt. Asama have provided us with high quality data. Employing these data, we were able to make clear the magma supply path beneath Mt. Asama shallower than 2 km below sea level based on the precise distribution of the hypocenters and the crustal deformation before and after the eruption on 1 September 2004.
 The Japan Meteorological Agency kindly provided us with the seismic data.