On 25 May 2009, a seismic event (mb 4.6) was recorded from a source in northeastern North Korea, close to the location of a previous seismic event on 9 October 2006. Both events have been declared to be nuclear tests by North Korea. For the more recent test, five seismo-acoustic arrays in South Korea recorded epicentral infrasonic signals. The signals are characterized by amplitudes from 0.16 to 0.35 microbar and dominant frequencies between 0.8 and 4.3 Hz. Celerities determined for the arrivals suggest that most of the infrasonic energy travelled as a stratospheric phase. Based on observed stratospheric amplitudes, the epicentral infrasonic energy was estimated to be equivalent to that expected from 3.0 tons of high explosives detonated on the surface. We conclude that this small energy estimate is due to the atmospheric coupling from the strong surface ground motion rather than the direct transfer of explosion energy to the air. This relatively small infrasonic to seismic energy ratio could be used to distinguish the event from a common surface explosion.
 Infrasound, seismic, hydroacoustic, and radionuclide monitoring techniques are part of the developing system for verifying the Comprehensive Nuclear-Test-Ban Treaty (CTBT) and provide the basis for characterizing possible nuclear explosions in the solid earth, the atmosphere and the oceans. To monitor compliance with the CTBT, the International Monitoring System global infrasound network is being deployed to detect one kiloton or greater atmospheric nuclear explosions (www.ctbto.org). Infrasound observations can record signals from natural and anthropogenic phenomena at local, regional and tele-infrasonic scales and can be used in the quantification of these other phenomena. The Korea Institute of Geoscience and Mineral Resources (KIGAM) and Southern Methodist University (SMU) have jointly operated seismo-acoustic arrays capable of observing infrasonic atmospheric pressure variations in and around the Korean Peninsula. Seven arrays are presently operational in Korea and have recorded data used to analyze the infrasound characteristics of near-surface explosions and strong earthquakes [Stump et al., 2004; Che et al., 2007].
 North Korea carried out its second underground nuclear test on 25 May 2009. Data collected by the dense seismic network in South Korea and neighboring countries were used to estimate the location of the explosion as 41.279 N, 129.065 E, approximately 2 km from the site of North Korea's 9 October 2006 previous nuclear test (J. S. Shin et al., Regional observations of the second North Korean nuclear test on the 25 May 2009, submitted to Geophysical Journal International, 2009). The regional seismic body wave magnitude was estimated to be 4.6 mb for the second test, which is larger than the 4.2 mb estimated for the first North Korean nuclear explosion. Based on standard magnitude and yield (W) relations, the nuclear explosion test yield is estimated to be approximately a kiloton in size. For example, the estimated seismic body wave magnitude and the magnitude-yield relationship for teleseismic body waves given by Murphy  of mb = 4.45 + 0.75 log W for fully tamped explosions in hard rocks gives a yield estimate of 1.6 kilotons.
 No distinct infrasonic signal associated with the 2006 test was recorded at the infrasound arrays in South Korea. However, five infrasound arrays in South Korea recorded epicentral infrasound signals from the larger second test. We analyzed the infrasonic signals associated with the second North Korean nuclear test and describe the characteristics of these regional signals. The progressive multi-channel correlation (PMCC) method [Cansi, 1995] was used to detect the coherent signals and derive the wave parameters. This method has proven to be effective in analyzing low-amplitude coherent waves within non-coherent noise [Le Pichon et al., 2002]. Ray tracing was performed to identify the observed infrasonic signals at each array and an azimuth intersection method was used to localize the infrasonic energy source. Wind corrected infrasound amplitudes were used to estimate the size of the infrasonic source based on high-explosives (HE) dataset for explosive sources detonated in the atmosphere [Whitaker et al., 2003].
2. Infrasound Arrays in South Korea
 KIGAM operates an infrasound network consisting of seven seismo-acoustic arrays in South Korea. Development of the arrays began in 1999, in collaboration with SMU, with the goal of detecting distant infrasound signals from natural and anthropogenic phenomena in and around the Korean Peninsula [Stump et al., 2004]. Figure 1 shows the locations of the infrasound arrays (solid hexagons) that observed infrasonic signals from the second nuclear test. This network includes the recently deployed seismo-acoustic arrays ULDAR and YAGAR. ULDAR is located on an island in ocean, 180 km from the mainland. This array, in particular, was deployed to enhance azimuth coverage of previous array distributions and to supply infrasound data recorded in an oceanic environment. All five arrays in Figure 1 have 1 km apertures with between 7 and 18 acoustic gauges. Individual array configurations and equipment have been designed to account for site specific conditions. Acoustic gauges at KSGAR, CHNAR, ULDAR, and BRDAR arrays are Chaparral Physics Model 2 microphones. The YAGAR array is equipped with Inter-Mountain Labs (IML) sensors. The infrasound sensors are connected to wind-noise reducers consisting of four to ten porous hoses arranged in a radial pattern. Data is sampled at 40 samples per second and are available for near real-time processing.
3. Infrasonic Signal Characteristics
 Seismic location techniques estimate a 00:54:42 (UTC) on 25 May 2009 origin time for the North Korean test. Following the regional seismic waves detected at densely distributed stations, infrasound signals generated in the epicentral region of the explosion began to arrive at the five distant infrasound arrays in Korea. The first infrasound phase arrived at KSGAR 1008 seconds after the seismic origin time. Infrasound phases followed at YAGAR, CHNAR, ULDAR, and BRDAR. However, no coherent and unambiguous signal was found on other two arrays (open hexagons) in Figure 1, due to higher noise level at KMPAR and due to longer propagation distance to TJIAR.
 PMCC was used to detect coherent infrasound signals at each array and estimate arrival times, azimuth and apparent velocity (calculated by WinPMCC v2.1 with the following parameters: window length of 20 s, frequency band of 0.5–5 Hz). Travel time for each signal was estimated based on the seismic origin time. Celerity (horizontal propagation velocity) was calculated by dividing the propagation range from the source by travel time. Figure 2 displays the output from PMCC for the infrasound data at each array where signals were detected. Subplots display the azimuths, apparent velocities of detected coherent signals, and waveform. Celerities and infrasonic phase identification based on the analysis are included in Figure 2.
 The arrays KSGAR, YAGAR, and BRDAR detected at least two infrasonic phases with different characteristics. Only one infrasonic phase was detected at CHNAR and ULDAR. Celerity and apparent velocity are important parameters for identifying the infrasound phase and explaining its propagation path in atmosphere. The measured celerities for all phases, except for CHNAR and the second phase at KSGAR were in the range of 281–310 m/s. This characteristic range corresponds to stratospheric infrasonic (Is) phases trapped between the stratopause and the ground [D. J. Brown et al., 2002]. The second arrival at KSGAR, with celerity of 254 m/s, corresponded to a thermospheric phase (It). This It phase was also characterized by a relatively low dominant frequency around 1.0 Hz as shown in Figure 2. In contrast, the dominant frequencies of the Is phases ranged from 2.6–4.3 Hz. The It apparent velocity of 375 m/s was higher than the range of 337–359 m/s for the Is phases and its corresponding higher turning height in the atmosphere. Although apparent velocity at CHNAR was 362 m/s which is slightly higher than the range of the Is phases, we considered this arrival as It based on its low celerity of 268 m/s. This phase also has low dominant frequency around 0.8 Hz. The signal durations of all phases were between 7 and 18 seconds. We measured the peak-to-peak values of largest pressures and found consistent features for elements of each array. Mean peak-to-peak amplitudes for the first Is phase were 0.19, 0.16, 0.35, and 0.30 microbar for KSGAR, YAGAR, ULDAR, and BRDAR, respectively.
4. Infrasonic Source Location and Energy
 Ray tracing was performed to identify the observed infrasonic phases for paths from the seismic source to each array. InfraMAP [Norris and Gibson, 2002] was used to perform these calculations with an atmospheric model (Horizontal Wind Model/Mass Spectrometer, Incoherent Scatter: HWM93,07 [Drob et al., 2008]/MSIS). Although Ground to Space (G2S) model is more sophisticated, it was not used for this study because of unavailability. Of the wind models, we used HWM93 for ray tracing because the model better explained our observations in phase identification for this particular event. Figure 3 displays possible eigen-rays predicted for each array and effective sound velocity structures used for ray tracings. Ray tracing with the atmospheric model predicts no arrivals for the two closest arrays to the source, KSGAR and YAGAR. In case of CHNAR and BRDAR, both Is and It phases were predicted but only the It phase was predicted for ULDAR. In particular, the stratospheric phase at BRDAR was included one reflection at the earth surface. Compared with observed phases in the previous chapter, however, there were discrepancies in phase types and celerities for each array. Thus, we interpret these discrepancies as reflecting the complexity and differences in atmospheric structures which could not be explained by the atmospheric model used in this study. The infrasonic energy source was also located by intersecting the azimuths of the first phases estimated at each array. For the most probable location, all intersections were weighted by the sine of the angle of intersection, and then a weighted average position for the most probable location of the event was determined [P. G. Brown et al., 2002]. The resulting infrasonic location was 15.7 km southwest from the reference seismic location.
 Based on the observations, first arrival phases at KSGAR, YAGAR, ULDAR, and BRDAR are attributed to stratospheric phases. As a result, the measured raw amplitudes (Praw) were corrected for the effects of stratospheric wind using Whitaker et al. . Our wind corrected amplitudes were then compared to wind corrected stratospheric observations from a set of atmospheric high-explosives (HE) tests of known explosive yield [Whitaker et al., 2003]. Stratospheric wind (Vs, m/s) from HWM93 at 50 km was used for the wind-corrected pressure amplitude (Pwca, microbar) following the relation Pwca = Infrasonic TNT energy equivalent (chgwt in kilotons) was then estimated based on Pwca = 59457(SR)−1.4072, where SR is the scaled range, R/ for distance R. The resultant average amplitude among the arrays was 0.29 (range from 0.16 to 0.64) microbar, which was then converted to an infrasonic source energy equivalent of 3.0 (0.7–8.6) tons of TNT. The infrasound-generating mechanism of the underground nuclear test of this study is different than that of the atmospheric explosions. But the comparison provides an assessment of the size of atmospheric explosion that would generate similar amplitude signals to those observed for this underground nuclear explosion. As may be expected, the contained nuclear explosion results in a significantly reduced infrasonic signal when compared to surface explosions of similar yield. The infrasonic amplitude is possibly reflecting the interaction of strong seismic waves with topographic highs overlying the nuclear explosion and in turn generating the infrasonic observations.
5. Conclusion and Discussion
 Epicentral infrasonic signals from an underground nuclear test by North Korea on 25 May 2009 were detected by an infrasound network in South Korea. Observed celerities and apparent velocities indicate that the observed signals propagated through stratospheric and thermospheric altitudes. The variations in celerities for the Is phases observed at the different arrays and discrepancies with results of ray tracings reflect the complexity of and differences in atmospheric structures which could not be explained by the standard atmospheric model used in this study. In addition, considering the strong absorption at thermosphere, the relatively high frequency of the It phases indicates real turning heights of this phase might be lower than that calculated in ray tracing. The inferred infrasonic location determined by intersecting the azimuths was offset about 15.7 km from the reference seismic location.
 When problematic seismic events occur, the foremost questions are whether the event resulted from an explosion or an earthquake and, if the former, the details of its size. The infrasound detections reported in this study do not provide conclusive evidence of a nuclear test. However, our analyses confirm that the observations are consistent with a near-surface explosion generating seismic and significantly reduced infrasonic signals. Although earthquakes with comparable or larger magnitudes to this event also generate epicentral infrasound, earthquakes characteristically have longer infrasonic signal durations including topographic effects which may differentiate them from surface explosions [Mutschlecner and Whitaker, 2005].
 We compared the infrasonic source energy from observed pressure amplitudes with an HE dataset and estimate the signal to be equivalent in strength with what is expected from a 3.0 ton HE explosion detonated at the surface. The region of the explosion is believed to consist of hard and competent rocks in an elevated mountainous area and a stable continental crust [Zhao et al., 2008]. We suggest that the infrasonic signals might have been produced when strong seismic waves generated by the explosion caused ground motion vibrations at topographic highs. Such infrasonic signals could also include contributions from permanent explosion caused topographic changes such as collapse and spall processes. The calculated infrasonic energy would then reflect only the energy that was released to the atmosphere via ground-to-air coupling. The large difference between the seismic and infrasonic energy release separates this event from near-surface explosions typically used in mining or engineering applications. Finally, these observations suggest that regional infrasound arrays can under some conditions detect signals expected from small atmospheric explosions at distances of several hundred kilometers.
 The authors thank B. Stump and C. Hayward for helpful discussions and support in the use of BBN Technologies' InfraMAP program during Il-Young Che's visiting scientist appointment at SMU. We also thank A. Le Pichon for support WinPMCC v2.1. This work was supported by the Basic Research Project of the Korea Institute of Geoscience and Mineral Resources funded by the Ministry of Knowledge Economy of Korea.