Traveling ionospheric disturbances generated by an epicentral ground/sea surface motion, ionospheric disturbances associated with Rayleigh-waves as well as post-seismic 4-minute monoperiodic atmospheric resonances and other-period atmospheric oscillations have been observed in large earthquakes. In addition, a giant tsunami after the subduction earthquake produces an ionospheric hole which is widely a sudden depletion of ionospheric total electron content (TEC) in the hundred kilometer scale and lasts for a few tens of minutes over the tsunami source area. The tsunamigenic ionospheric hole detected by the TEC measurement with Global Position System (GPS) was found in the 2011 M9.0 off the Pacific coast of Tohoku, the 2010 M8.8 Chile, and the 2004 M9.1 Sumatra earthquakes. This occurs because plasma is descending at the lower thermosphere where the recombination of ions and electrons is high through the meter-scale downwelling of sea surface at the tsunami source area, and is highly depleted due to the chemical processes.
 Ionosphere is disturbed by large earthquakes [Ducic et al., 2003; Liu et al., 2006a] and tsunamis [Artru et al., 2005; Liu et al., 2006b; Rolland et al., 2010]. When a vertical sudden displacement of the ground and sea surface caused by the earthquake and tsunami excites acoustic-gravity waves in the atmosphere [seeWatada, 2009], the acoustic-gravity waves propagate into the ionosphere and disturb ionospheric plasma. The disturbances originated from the ground and sea surface motions were observed initially with ionosondes and high-frequency (HF) Doppler sounding systems [e.g.,Leonard and Barnes, 1965] and recently with the measurement of Ground Positioning System - Total Electron Content (GPS-TEC) [e.g.,Ducic et al. 2003]. As an extraordinary case, the M9.1 Sumatra-Andaman earthquake (Sumatra EQ), which occurred at 00:58 (Universal time: UT) on 26 December 2004, caused large-scale ionospheric disturbances. The sea surface displacement associated with tsunami generated a traveling ionospheric disturbance with several periods [Heki et al., 2006; Liu et al., 2006b]. Rayleigh waves travelling which produces the large vertical displacement also generated ionospheric disturbances [Liu et al., 2006a]. Post-seismic 4-minute monoperiodic atmospheric resonance, i.e., acoustic-gravity wave oscillations between the lithosphere and the thermosphere was observed during four hours, one hour after the mainshock [Choosakul et al., 2009]. As a similar event, the M8.8 Chile EQ, which was a large subduction EQ, occurred at 6:34 UT on 27 February 2010. A tsunami was also generated in the Chile EQ and the tsunami that traveled across the Pacific Ocean caused ionospheric disturbances [Galvan et al., 2011]. The M9.0 off the Pacific coast of Tohoku earthquake (Tohoku EQ) which was also a large subduction EQ occurred at 5:46 UT on 11 March 2011 in the western Pacific Ocean. The epicenter was located in 38.297°N and 142.372°E and the focal depth was 30 km. The Tohoku EQ resembles the Sumatra EQ in its magnitude and fault type. Liu et al. reported that the similar large-scale seismogenic and tsunamigenic ionospheric disturbances in the Tohoku EQ. As shown later, an ionospheric hole, widely a sudden depletion of TEC, is found. The ionospheric hole is attributed to the Tohoku EQ and is different from the above mentioned co-/post-seismic ionospheric phenomena. The ionospheric hole occurs after the acoustic-gravity waves generated by the EQ reached the ionosphere and gradually disappeared within a few tens of minutes. On the other hand, the ionospheric holes did not appear in the inland EQs such as the 1999 M7.6 Taiwan Chi-Chi inland EQ [seeLiu et al., 2010, Figure 3]. Therefore, the ionospheric hole is expected to be attributed to a tsunami.
 In this paper, we present that the ionospheric hole appears in the subduction earthquake and is generated by the tsunami, verifying the results of not only the 2011 M9.0 Tohoku but also the 2004 M9.1 Sumatra and the 2010 M8.8 Chile EQs. The physical mechanism of the tsunamigeneic ionospheric hole is discussed.
 The GPS is used to derive the TEC along the slant path between the satellite and the receiver using dual frequency (1575.42 and 1222.60 MHz) radio signals. Assuming that the ionosphere exists at 300 km as a thin layer, vertical TEC is obtained with considering the elevation angle of the GPS satellites [Liu et al., 1996]. Since there are several factors such as the satellite and receiver instrumental biases, we focus on the variation of slant TEC. The point that the ray path from a GPS satellite to a ground-based receiver intercepts the ionosphere is named ionospheric point. Each ionospheric point which acts as a monitoring station by measuring the TEC is employed to detect the ionospheric disturbances. Footprint of the ionospheric point on Earth's surface is termed a subionospheric point (SIP).
 In this study, GPS data are provided by the Geographical Survey Institute (GSI) of Japan which has installed more than 1000 receivers in Japan as a nationwide GPS array, named GPS Earth Observation Network (GEONET) (ftp://terras.gsi.go.jp/). The sampling time of the GPS data record in GEONET is 30 seconds. For the Chile EQ case, the GPS data are provided by Instituto Geografico Nacional De La República Argentina (http://www.ign.gob.ar/DescargaRamsac) for AZUL, CFAG, CSJ1, CSLO, LHCL, MZAC, UNSJ, SL01, UCOR, SVIC, TERO, JBAL, and ALUM stations (15-second sampling time) and the Scripps Orbit and Permanent Array Center (SOPAC) (http://sopac.ucsd.edu/) for COPO and ANTC (30-second sampling time). For the Sumatra EQ case, the GPS data are also provided by SOPAC for SAMP and NTUS (30-second sampling time), and ABGS, MKMK, NGNG, and PRKB (2-minnute sampling time).
 For the comparison between the ionospheric disturbance and the tsunami generation, the tsunami wave in the Tohoku EQ is calculated. Tsunamis generated by the Tohoku EQ are numerically computed from the slip distribution estimated from the tsunami waveforms and the coseismic crustal deformation data [Tanioka et al., 2011]. The linear shallow water equations with a spherical coordinate system are numerically solved using the finite difference method as described by Johnson . Because the tsunamis on the deep ocean are not much affected by nonlinear coastal effects, the simulation of the tsunami using the linear shallow water equations is widely accepted [Synolakis et al., 2008]. The grid spacing used in the tsunami simulation is 90 arc-seconds. The grid bathymetry data for the tsunami simulation are made from the General Bathymetric Chart of the Oceans (GEBCO) 30 arc-second data set and the Japan Hydrographic Association's M7001 and M7002 bathymetric contour data sets. The computation area ranges from 130°E to 160°E and from 10°N to 50°N.
3. Result and Discussion
 Several visible satellites above the epicentral region were observed around the time of mainshocks of the EQs. Figure 1illustrates the epicenter, the location of receiving stations, and the SIPs together with the slant TEC time-series of one of the visible satellites around the mainshocks with the reference slant TEC curves, which are obtained by the similar orbits of the same satellite one day later. Dst indices at the time of the mainshock of the EQ and the reference days are about −80 and −44 nT in the Tohoku EQ, −5 and −3 in the Sumatra EQ, −1 and 16 in the Chile EQ, respectively (see World Data Centers for Geomagnetism, Kyoto University;http://wdc.kugi.kyoto-u.ac.jp/dstdir/index.html). Therefore, the Tohoku EQ occurred during the recovery phase of the magnetic storm, while the Sumatra and the Chile EQs occurred under the geomagnetically quiet conditions. Even in the Tohoku EQ, the slant TECs on the EQ and the reference days reveal the similar tendencies. In the present study, both the slant TEC curves of the EQ and the reference days are offset to be zero at the time of the mainshock. From the two curves of the receiving stations, the following features arise in the three large EQs: 1) Ionospheric disturbance occurs around 9 minutes after the mainshock. 2) In some of the receiving stations, the enhancement of TEC with the amplitude of a few TECu (1 TECu = 1 × 1016 electron / m2) is initially observed (type A), while in the others, the TEC starts decreasing (type B). 3) Then, the large depletion of TEC occurs within a few minutes. The amplitude of the large depletion exceeds about 5 TECu and is large near the epicenter. The depletion with four-minute monoperiodic signature lasts for tens of minutes. InFigure 1c, only the SIP for SAMP (satellite 23) is located around the tsunami source region [see Tanioka et al., 2006, Figure 3], while the others are far from this region. Therefore, the large depletion of TEC is observed only at SAMP in Figure 1c. The result of the Sumatra EQ also is consistent with that of the Tohoku earthquake.
 In the Tohoku EQ, the spatial distributions of slant TEC difference between the EQ and the reference days and simulated the tsunami height are shown in Figure 2. The first panel of Figure 2adisplays the slant TEC differences at the times when the mainshock occurs. After the mainshock, it is recognized that the wavefront of acoustic-gravity generated by the EQ reaches the ionosphere (second panel ofFigure 2a), and then the acoustic-gravity waves slightly propagate in the ionosphere (third panel ofFigure 2a). Finally, the large depletion of plasma widely appears (the fourth panel of Figure 2a). Since the large depletion is localized just above the tsunami source area, the ionospheric hole is highly expected to be attributed to the tsunami. After the initial enhancement of TEC appeared (the second panel of Figure 2a), the enhancement of TEC spatially enlarges and travels southwestward, and small depletion appears northeastward (the third panel of Figure 2a). The enlarged enhancement and the depletion of the slant TEC differences correspond to the time-series of types A and B (see the right bottom of the third panel ofFigure 2a). The ray tracing calculation of the acoustic waves generated by the EQ shows that the acoustic waves horizontally propagate when they reach the ionosphere [Heki and Ping, 2005]. Since the plasma moves along the magnetic field line, the horizontal acoustic-gravity wave propagation in the southwest and the northeast region causes the plasma migrate upward and downward, respectively. This might be the reason of the discrepancy of the slant TEC difference in the types A and B (the third panel ofFigure 2c). After that, the large depletion of the slant TEC difference widely occurs above the tsunami source area. Coincidently, the wavefront of the acoustic-gravity wave further travels southwestward with approximately 1.8 km/s (seeFigure 3a), while it is not clearly seen northeastward because the plasma moves along the magnetic field line as discussed in Heki and Ping  as well. On the other hands, Figure 2b shows two snapshots of the tsunami propagation due to the Tohoku EQ 1 minute and 15 minutes after the mainshock, which approximately correspond to the times of the second and the fourth panels of Figure 2a. As shown in the second panel of Figure 2b, the downwelling of the sea surface after the tsunami propagation occurs, and then the meter-scale downwelling of the sea surface yields the large downwelling of thermosphere with the plasma in the scale of tens of kilometers, which finally produces the ionospheric hole.
Figure 3shows the time-series of the slant TEC difference and the simulated tsunami area. The change of the slant TEC difference starts 9 minutes after the mainshock, which is the time when the acoustic-gravity wave reaches the ionosphere (seeFigure 4a), and then the disturbance propagates with approximately 1.8 km/s. Meanwhile, the wavefront of tsunami propagates with 260 m/s from the tsunami source area (Figure 3c). These different velocities imply that the initial slant TEC variation is caused by the acoustic-gravity wave excited at the tsunami source area, not the wave front of tsunami. Note that the comparison of these velocities is slightly imprecise because the different directions of the initial slant TEC variation and the tsunami discussed here are taken. After the initial slant TEC variation (Figure 3a), the large-scale electron depletion, namely the ionospheric hole, occurs. While the propagation disturbances following the initial TEC variation are observed (Figure 3a), the large-scale depletion is located only over the tsunami source area and does not propagate. Theses imply that only the initial downwelling of tsunami produces the ionospheric hole and the propagation of tsunami downwelling does not produce it. Since the sudden meter-scale downwelling of the sea surface yields large the downwelling of the thermosphere with the plasma in the scale of tens of kilometers (seeFigure 4b), the large downwelling of ionospheric plasma causes the plasma decrease due to the high recombination rate at the lower thermosphere through the chemical reactions, and then the ionospheric hole is generated. Although the proposed mechanism of the tsunamigenic ionospheric hole is qualitative, the numerical simulation will be investigated in the future in the view of ocean-atmosphere-ionosphere coupling.
 High-pass filtered TEC data (e.g., the time windows of 0.5 to several minutes) has been frequently used to extract the variation of ionospheric disturbance generated by the EQs and the tsunamis [e.g.,Liu et al., 2011]. The lifetime of the tsunamigenic ionospheric hole is in the order of tens of minutes and is much longer than that of the other ionospheric disturbances. Therefore, the high-pass filtered TEC data is not suitable to observe the tsunamigenic ionospheric hole, and slant or vertical TEC data without the high-pass filter should be used.
 We emphasize that inland large EQs do not produce the ionospheric hole. As introduced before, Liu et al. reported the co-seismic ionospheric disturbance in the 1999 M7.6 Chi-Chi inland EQ, Taiwan. Their slant TEC time-series [Liu et al., 2010, Figure 3] show that after the initial TEC enhancement appeared, the depletion with the similar amplitude and period happens without the ionospheric hole. Furthermore, in the case of the 2008 M7.9 Wenchuan EQ, the recent largest inland EQ in the world, no paper that shows the slant TEC time-series without the high-pass filter has been published. However, the ionospheric hole may not exist, because the high-pass filtered data including the ionospheric hole shows a pre-seismic enhancement of high-pass filtered TEC which was not observed according toAfraimovich et al. .
 Besides traveling ionospheric disturbances generated by an epicentral ground/sea surface motion, ionospheric disturbances associated with Rayleigh-waves as well as post-seismic 4-minutes monoperiodic atmospheric resonances and other-period atmospheric oscillations [Saito et al., 2011; Tsugawa et al., 2011; Rolland et al., 2011; Kamogawa et al., 2012], the ionospheric hole appears due to the tsunami induced by the subduction EQs. The tsunamigenic ionospheric hole is caused by the meter-scale downwelling of the sea surface at the tsunami source area yields the hundred-kilometer-scale ionospheric plasma downwelling and plasma depletion due to the high recombination of plasma at the lower thermosphere through the chemical processes.
 Yoshihiro Kakinami and Masashi Kamogawa are co-first authors. We thank K. Heki (Hokkaido Univ.), H. Kanamori (Caltech), M. Hashimoto (Kyoto Univ.), and P. Wu (Japan Agency for Marine-Earth Science and Technology) for their comments and discussion. GPS data in this study were provided by the Geographical Survey Institute of Japan and National Research Institute for Earth Science, Instituto Geografico Nacional De La Republica Argentina, and the Scripps Orbit and Permanent Array Center. This research was partially supported by the Ministry of Education, Science, Sports and Culture, Grant-in-Aid for Young Scientists (B), 21710180, 2009 (M.K), and Scientific Research (C), 20510171, 2008 (M.K.), Observation and Research Program for Prediction of Earthquakes and Volcanic Eruptions, 2009 (M.K. and Y.K.), Heiwa Nakajima Foundation, 2011 (M.K, Y. K., and J.Y.L.), the National Science Council project NSC 98-2116-M-008-006-MY3 grant of the National Central University (Y.K. and J.Y.L.), and Earth Observation Research Center, Japan Aerospace Exploration Agency (Y.K. and S.W.).
 The Editor thanks two anonymous reviewers for their assistance evaluating this paper.