Large earthquakes that occurred in the Sumatra region in 2004 and 2005 generated acoustic waves recorded by the Diego Garcia infrasound array. The Progressive Multi-Channel Correlation (PMCC) analysis is performed to detect the seismic and infrasound signals associated with these events. The study is completed by an inverse location procedure that permitted reconstruction of the source location of the infrasonic waves. The results show that ground motion near the epicenter and vibrations of nearby land masses efficiently produced infrasound. The analysis also reveals unique evidence of long period pressure waves from the tsunami earthquake (M9.0) of December 26, 2004.
 Sources such as atmospheric or buried explosions and shallow earthquakes are known to produce infrasonic waves. Pressure waves from the sudden strong vertical ground displacements have been detected during the Alaskan earthquake of March 27, 1964, at distances of thousands kilometers from the origin [Mikumo, 1968; Young and Greene, 1982]. Cook  reported on pressure waves radiated locally by seismic waves during the Montana earthquake on August 18, 1959. Distinct mechanisms of pressure wave generation during large earthquakes have been identified: (i) pressure changes due to the vertical displacement of the seismic waves near the infrasound station (local infrasound). These receptions are associated in part with the instrumental response of the microbarometer to seismic waves [Bedard, 1971; Kim et al., 2004; Starovoit and Martysevich, 2005; Alcoverro et al., 2005], (ii) ground-coupled air waves generated near the epicenter region due to an integrated effect of the sound radiated by the ground motion [Olson et al., 2003; Takahashi et al., 1994], and (iii) infrasonic waves radiated by the topography when seismic surface waves travel through mountainous regions [Le Pichon et al., 2003; Mutschlecner and Whitaker, 2005].
 Infrasound arrays in the Pacific and Indian Oceans that are part of the International Monitoring System (IMS) recorded three distinct waveform signatures associated with the December 26, 2004 Sumatra earthquake (3.31°N–95.95°E at 00:58:53 UTC, M9.0 (USGS Earthquake Hazards Program, http://earthquake.usgs.gov/; the authors are aware of revised magnitude estimates for these events, which are outside the scope of this paper): (i) seismic arrivals (primary and surface waves) from the earthquake, (ii) tertiary arrivals (T-phases) propagated along the SOund Fixing And Ranging channel (SOFAR) in the ocean [Tolstoy and Ewing, 1950], and coupled back to the ground, and (iii) infrasonic arrivals generated by seismic wave-induced ground motion [Garcés et al., 2005a, 2005b; Le Pichon et al., 2005]. The seismic and T-phase recordings are due in part to the sensitivity of the microbarometers to ground vibration, whereas the infrasound arrivals correspond to dispersed acoustic waves propagated through atmospheric waveguides. Similar recordings were also observed during the March 28, 2005 Northern Sumatra earthquake (2.07°N–97.01°E at 16:09:36, M8.7) and the April 10, 2005 Mentawai earthquakes (1.67°S–99.62°E at 10:29:11, M6.7 and 1.80°S–99.57°N at 11:14:19, M6.5). In this paper, detailed analyses of these signals are presented. We compare the location of the main sources of infrasound and discuss the different source mechanisms involved in these events.
2. Infrasound Measurements
 The IS52 infrasound station (Diego Garcia, 7.38°S–72.48°E) is located in the British Indian Ocean Territory, 1560 km south of the tip of India. This station is composed of seven MB2000 type microbarometers that can measure pressure fluctuation from 0.003 up to 27 Hz. IS52 recorded coherent infrasonic waves from several large earthquakes in the Sumatra region that occurred at ∼3000 km away from the station. The wave parameters are calculated with the Progressive Multi-Channel Correlation method (PMCC) [Cansi, 1995].
 The detection results for the December 26, 2004 earthquake are divided in four groups (Figure 1a). Group I is related to seismic waves with a horizontal trace velocity generally greater than 3 km/s. These arrivals are primarily a manifestation of the seismic response of the MB2000 microbarometer. In the processed frequency band (0.02–0.15 Hz), values of horizontal trace velocity are consistent with the propagation of seismic waves. From 03:00 to 03:30, group II indicates coherent infrasonic waves with a mean trace velocity of 0.35–0.36 km/s, a dominant period of ∼10 s, and a peak-to-peak amplitude of ∼0.5 Pa. They are associated with an extended radiation area near the epicenter region. From 03:40 to 07:00, group III corresponds to large coherent infrasonic waves with similar values of trace velocity, a period of maximum amplitude around 30 s, and a peak-to-peak amplitude of ∼2 Pa. The PMCC analysis displays a clear backazimuth trend of over 40°, starting near the northern tip of Sumatra and ending in the northern margin of Bay of Bengal. Group IV overlaps group III. It indicates a second coherent infrasound wave train along a bearing of ∼60°. Its arrival time is consistent with infrasound generated near the epicenter by the first M7.3 aftershock (6.88°N–92.94°E at 04:21:29). Moreover, other infrasonic signals are also detected: (i) microbaroms with a bearing of 125° and frequency above 0.1 Hz, (ii) Mountain Associated Waves (MAW) from the Himalaya with periods larger than 20 s and a bearing of ∼15°.
 Coherent signals generated by the March 28, 2005 earthquake are also detected. Figure 1b shows two groups of arrivals. In group I, from 16:10 to 17:20, velocities of 2–3 km/s correspond to the detection of seismic surface waves. In group II, coherent infrasonic waves are detected from 18:45 to 19:45 with a mean azimuth of 74°. As opposed to the December 26, 2004 earthquake, no clear azimuthal variation is observed.
Figure 1c presents the detection results of the M6.7 and M6.5 April 10, 2005 Mentawai earthquakes. In the frequency band processed (0.1–2 Hz), the seismic arrivals are not detected. Although they are not prominent, they can be observed at lower frequency. Two similar infrasound wave trains associated with these earthquakes are observed with a dominant period of ∼1 s and a mean azimuth of 78°. Each of them consists of two distinct arrivals.
3. Spatial Distribution of Infrasound Sources
 From these observations, the main source regions of infrasound can be reconstructed. The input parameters of the location procedure include the measured azimuths and arrival times, the origin time and coordinates of the epicenter. The velocity models used describe the propagation of the seismic surface waves and the tsunami from the epicenter, and also propagation of infrasound through the atmosphere in the direction of IS52. For the December 26, 2004 earthquake, we use a velocity of 3.3 km/s for the seismic surface waves, and a speed related to the square root of the water depth for the tsunami waves (Groups II and III in Figure 1a, respectively). The March 28 and April 10, 2005 earthquakes generated a small tsunami that did not produce detectable infrasound. For those events, seismic surface waves were the dominant source of infrasound. The atmospheric part is described by sound velocity and wind speed profiles provided by the time-varying Ground to Space (G2S) atmospheric model [Drob et al., 2003] specific to the location and time of the events. Infrasonic wave propagation is performed using a 3D ray theory modeling. The equations used include the spatio-temporal variations of horizontal wind terms along the ray paths [Dessa et al., 2005]. For these events, two dominant wave guides are predicted: (i) thermospheric phases refracted below 120 km (slowness below 2.75 s/km), (ii) stratospheric ducted waves refracted below 45 km (slowness above 2.75 s/km). The slowness distribution derived from the measured trace velocity presents a maximum between 2.73 and 2.89 s/km. These values correspond to infrasonic waves propagating in the stratospheric duct with a celerity of 0.30 ± 0.01 km/s. The component of the wind transverse to the propagation direction deflects the rays from the original launch direction. The measured azimuths are wind-corrected by a relatively small deviation (less than 2°). Taking into account uncertainties in the measured azimuths and the velocity models, a maximum location error of ∼100 km is estimated.
 For the December 26, 2005 earthquake, the long signal duration and the large azimuth variations observed in the [0.02–0.15] Hz frequency band, suggest that wide regions acted as sources of infrasound (Figure 2). The southern part of the reconstructed source regions falls into line with the mountains of the extreme north of Sumatra. The source distribution is also concentrated above the continental shelf along the coast of the Bay of Bengal. For the March 28, 2005 earthquake, the radiating zone is limited to a smaller region (Figure 3a, ∼2° in latitude and longitude). The source distribution points to the ∼2900 m Barinson Mounts on island of Sumatra. The maximum of the distribution is found at ∼150 km to the south-east of the epicenter. On April 10, 2005, infrasound measurements and location results are comparable for the M6.7 and M6.5 earthquakes. Figure 3c presents the location results for the M6.5 earthquake. Two peaks in the source distribution are found. They are associated with the two distinct arrivals observed in the recordings (Figures 1c). One region is located over the Siberut Island, the other one near the ∼3800 m Kerindji mount on the Island of Sumatra.
 In order to check the consistency of regions radiating infrasound with areas of strong ground motion, we use the slip patches model developed by Bouchon [Vallée and Bouchon, 2004] which looks for the simplest extended source model able to explain the teleseismic seismograms of the March 28 and April 10 earthquakes. The first and second order characteristics of the event (location, depth, duration, focal mechanism, and refined kinematic parameters such as spatial slip distribution on the fault and rupture velocity) are calculated from teleseismic P and SH body waves. Then, from the model of the fault rupture, synthetic seismograms are computed on a grid in the vicinity of the epicenter using a discrete wave number method [Bouchon, 1981].
4.1. Infrasound From the Epicenter and Mountainous Regions
 Seismic source studies of the December 26, 2004 mega-earthquake showed the existence of a big slip patch off the west coast of Banda Aceh (Y. Yagi, Preliminary results of rupture process for 2004 of northern Sumatra giant earthquake, International Institute of Seismology and Earthquake Enginnering (IISEE), http://iisee.kenken.go.jp/staff/yagi/eq/Sumatra2004/Sumatra2004.html, 2005a, hereinafter referred to as Yagi, 2005a) which may explain the infrasound source regions in the northern part of Sumatra (Figure 2). A classical subduction rupture occurred during the March 28 event. The rupture propagated bilaterally, north-westward along a segment ∼100 km, and ∼200 km southeast-ward. The calculated focal mechanism is consistent with Harvard CMT moment tensor solution and the slip distribution provided by Y. Yagi (Preliminary results of rupture process for the March 28, 2005 of northern Sumatra, Indonesia earthquake, International Institute of Seismology and Earthquake Enginnering (IISEE), http://iisee.kenken.go.jp/staff/yagi/eq/Sumatra2005/Sumatra20050328.html, 2005b). The maximum slip is located to the southeast part of the epicenter (Figure 3b), near the infrasound radiating zone shown in Figure 3a. Compared to the March event, the source mechanism for the M6.7 April 10 earthquake indicates a steeper dip (∼60°) and a smaller rupture length (∼40 km). The main slip is found eastward and westward of the fault (Figure 3d) with a maximum of ground displacement near the infrasound source regions (Figure 3c). For the March 28 and April 10 events, these results confirm that most of the infrasound energy is radiated by the vibration of large islands near the epicenter, and by seismic surface waves traveling through nearby high mountain ranges.
4.2. Infrasound From Tsunami
 Following the December 26, 2004 earthquake, a radiating zone of ∼1500 km length in the north-northeastern part of the Bay of Bengal has been reconstructed. No similar result has been previously reported. These location results provide unique evidence that the tsunami generated infrasonic waves. Different source mechanisms for infrasound generated by the tsunami waves are proposed:
 • The tsunami slows down with growing wave height as it travels into shallower water and collapses into a series of breaking waves. As the water depth decreases, the wavelength of the tsunami decreases, reaching values of few tens of kilometers, comparable with the dominant period (∼30 s) of the infrasonic signals. Due to the compression and expansion of large volumes of air suddenly released, acoustic waves whose amplitude depends on the height and length of the ocean waves may be produced.
 • The interaction of the tsunami waves with the shoreline. Alternately, the source process may involve multiple reflections of the tsunami energy from steep bathymetry, leading to resonant triad interactions as encountered in open-sea swells [Willis et al., 2004].It appears that the observed infrasound source region does not reflect the amplitude of the tsunami waves. Infrasound was not observed from the eastern coast of Sri Lanka, which is surprising considering the amplitude of the tsunami in this area. The tsunami's interaction with the continental shelf (bathymetry and extent), and a strong directionality along the tsunami propagation direction during the energy transfer could explain the lack of detection from the eastern coast of India in Diego Garcia.
5. Concluding Remarks
 Large earthquakes in the Sumatra region (M > 6.5) generated infrasonic waves which have been observed at several thousands of kilometers. For such events, infrasonic measurements are valuable for the analysis of the remote effects of earthquakes. The observed azimuth variations and the long signal duration are explained by distant source regions radiating infrasound. Two main mechanisms are identified: (i) the violent ground motion of islands or ocean surface near the epicenter, (ii) the increase of the effective infrasound source region when the seismic surface waves travel through high mountains where radiation occurs. The consistency between the source locations and mountainous regions provides a means to validate the infrasonic celerity model derived from the G2S wind profiles with an uncertainty less than 10 m/s. The December 26, 2004 earthquake generated large coherent waves with a dominant period of ∼30 s over four hours. The location results indicate that the tsunami also acted as sources of infrasonic waves at lower frequencies when it propagated in shallow water. Future work could explore the relation between the combined effects of the bathymetry and the geographic configuration of the coastlines, and the long period pressure waves from the tsunami's interaction with the continental shelf. It is hoped that joint studies, combining propagation modeling and seismo-hydro-acoustic measurements will allow a better understanding of the physical mechanism involved in the generation of infrasound from large submarine earthquakes.
 The authors are grateful to Y. Cansi and J. Guilbert for their interest in this study and the helpful discussions we had during the completion of this work. Many thanks also to H. Hebert for providing us simulation results related to the propagation of the tsunami. The views expressed herein are those of the authors and do not reflect the views of the CTBTO Preparatory Commission.