Three large-scale infrasound calibration experiments were conducted in 2009 and 2011 to test the International Monitoring System (IMS) infrasound network and provide ground truth data for infrasound propagation studies. Here we provide an overview of the deployment, detonation, atmospheric specifications, infrasound array observations, and propagation modeling for the experiments. The experiments at the Sayarim Military Range, Israel, had equivalent TNT yields of 96.0, 7.4, and 76.8 t of explosives on 26 August 2009, 24 January 2011, and 26 January 2011, respectively. Successful international collaboration resulted in the deployment of numerous portable infrasound arrays in the region to supplement the IMS network and increase station density. Infrasound from the detonations is detected out to ~3500 km to the northwest in 2009 and ~6300 km to the northeast in 2011, reflecting the highly anisotropic nature of long-range infrasound propagation. For 2009, the moderately strong stratospheric wind jet results in a well-predicted set of arrivals at numerous arrays to the west-northwest. A second set of arrivals is also apparent, with low celerities and high frequencies. These arrivals are not predicted by the propagation modeling and result from unresolved atmospheric features. Strong eastward tropospheric winds (up to ~70 m/s) in 2011 produce high-amplitude tropospheric arrivals recorded out to >1000 km to the east. Significant eastward stratospheric winds (up to ~80 m/s) in 2011 generate numerous stratospheric arrivals and permit the long-range detection (i.e., >1000 km). No detections are made in directions opposite the tropospheric and stratospheric wind jets for any of the explosions. Comparison of predicted transmission loss and observed infrasound arrivals gives qualitative agreement. Propagation modeling for the 2011 experiments predicts lower transmission loss in the direction of the downwind propagation compared to the 2009 experiment, consistent with the greater detection distance. Observations also suggest a more northerly component to the stratospheric winds for the 2009 experiment and less upper atmosphere attenuation. The Sayarim infrasound calibration experiments clearly demonstrate the complexity and variability of the atmosphere, and underscore the utility of large-scale calibration experiments with dense networks for better understanding infrasound propagation and detection. Additionally, they provide a rich data set for future scientific research.
 The International Monitoring System (IMS) of the Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO) is a global network consisting of seismic, hydroacoustic, infrasound, and radionuclide stations built to monitor for nuclear test explosions. Currently 45 of 60 infrasound arrays have been completed and certified into the IMS. The CTBT itself makes no reference to explosion magnitude, but for practical reasons the IMS infrasound network was designed to detect and locate atmospheric explosions of at least 1 kt TNT-equivalent yields with at least two stations [Christie and Campus, 2010]. The development of the IMS infrasound network has helped stimulate research and development in the field of infrasound. Numerous studies have shown that the IMS network routinely detects large infrasonic events at multiple stations, such as bolides [Silber et al., 2011], chemical explosions [Ceranna et al., 2009], large volcanic eruptions [Dabrowa et al., 2011], earthquakes [Le Pichon et al., 2006], tsunamis [Le Pichon et al., 2005], and microbaroms [Garces et al., 2004]. Several studies have modeled the detection capability of the global IMS infrasound network. Le Pichon et al.  focused on the influence of the stratospheric winds and found that the majority of IMS infrasound detections are in the downwind stratospheric wind direction. They estimate that the IMS network is capable of detecting a ~500 t explosion at any time of the year with at least two stations, but that much smaller explosions (e.g., 50 t yields) could be detected under favorable propagation conditions. Green and Bowers  used a probabilistic model accounting for stratospheric winds and found that chemical explosions >210 t will be detected at over 95% of the Earth's surface at any time of the year by at least two stations of the full IMS network.
 The middle and upper atmosphere is dynamic and difficult to sample evenly and consistently, contributing to large uncertainties in long-range acoustic propagation modeling and often resulting in less than satisfactory agreement between modeling and observations [Norris et al., 2010]. In addition, the relatively low station density of the IMS (~2000 km between stations) only permits coarse comparisons between modeling and observations. Higher station density allows a more detailed view of acoustic propagation and atmospheric structure. Large anthropogenic sources such as chemical explosions [e.g., Ceranna et al., 2009] and natural events such as volcanic explosions [e.g., Matoza et al., 2011] have provided more detailed studies of the atmosphere and signal identification. However, these events are somewhat rare and difficult to predict, thereby complicating installation of higher-density networks to validate acoustic propagation modeling and middle- and upper-atmospheric specifications. Furthermore, accurate source constraints, such as explosive yield and source waveforms, are difficult to impossible to obtain for unexpected events. Some previous large-scale infrasound calibration experiments have taken place [Herrin et al., 2008], using rocket-launched low-yield explosions in the middle atmosphere. The use of ground-coupled airwaves on seismic networks has led to studies with higher station density [e.g., Hedlin et al., 2010], but these studies are limited to travel time information alone. Therefore, it is crucial to have a large, well-constrained source and high-density infrasound network to fully test the IMS network and atmospheric propagation models.
 In 2009 and 2011, three large-scale infrasound calibration experiments were performed. Each calibration experiment consisted of a singular surface detonation of chemical explosives with charge weights of 82, 10.24, and 102.08 t [Gitterman et al., 2011] on 26 August 2009, 24 January 2011, and 26 January 2011, respectively, at the Sayarim Military Range, Israel (Table 1). These explosions produced significant infrasound detected by numerous permanent and temporary infrasound arrays deployed across the region. The goal of the calibration experiments was to test the IMS infrasound network and evaluate atmospheric specifications and propagation models using large, well-constrained infrasound sources and infrasound networks of unprecedented station density. This manuscript provides an overview of the calibration experiments, including deployment (section 2), data processing and methods (section 3), seasonal winds and atmospheric specifications (section 4), explosion characterization (section 5), infrasonic signal detection and identification (section 6), and a discussion of the results and implications (section 7). The Sayarim calibration experiments represent a considerable and successful collaboration between the CTBTO and other international groups and will provide a rich ground truth data set for more detailed infrasound studies in the future.
Table 1. Explosion Characteristics
Yield (t TNT equivalent)
26 August 2009
24 January 2011
26 January 2011
 For the calibration experiments, numerous pressure gauges and temporary infrasound arrays were deployed to supplement the existing IMS infrasound network. Arrays were deployed to the north and west of the source (Figure 1a) for the 2009 experiment, due to the predicted summer westward stratospheric winds expected to promote long-range propagation. IMS arrays in the region include IS26 (Germany), IS31 (Kazakhstan), IS43 (Russia), and IS48 (Tunisia). Fifteen temporary arrays were deployed in total. The temporary arrays were deployed by numerous groups and consisted of variable array configurations and equipment summarized in Table 2. A five-element infrasound array (IMA) was deployed by the National Data Center of Israel (Soreq) in northern Israel, Mt. Meron (colocated with IMS seismic array MMAI). The University of Mississippi National Center for Physical Acoustics (NCPA) and University of Alaska Fairbanks (UAF) deployed sensors in Israel as well. A linear array of three sensors was deployed by the NCPA and Geological Survey of Cyprus in Cyprus (CYPR). The University of Hawaii Infrasound Lab and National Observatory of Athens deployed three arrays in Greece: Rhodes (RHOD), Crete (CRET), and the Peloponnese (PELO). The University of Florence deployed one array in Southern Italy (CALA), and the CTBTO deployed arrays in North Italy (I62) and Austria (I63). In France, the French Alternative Energies and Atomic Energy Commission (CEA/DASE) deployed an array in Provence (PROV), as well as operating research arrays near Paris (CEA) and Flers (FLERS). In Netherlands, the Royal Netherlands Meteorological Institute (KNMI) operates three research arrays in the region (DBN, DIA, and TEX). All arrays consisted of at least three sensors with flat (−3 dB) frequency responses in the band of interest (0.1–20 Hz) and were sampled at >20 Hz. The majority of the sensors were either Chaparral Physics Model 25 or Martec Tekelec MB2005 [Ponceau and Bosca, 2010]. Coordination for the 2009 temporary deployments was led by Dr. Milton Garces of the University of Hawaii and sponsored by the U.S. Army Space and Missile Defense Command (SMDC).
Table 2. List of Stations for 2009 Experiment as a Function of Distance From the Source
Back Azimuth (°N)
MB2000 x 5
NCPA IS Mic x 3
Chaparral 2.2 x 6
Chaparral 2.2 x 6
Chaparral 2.2 x 6
MBAR-D-4V – x 6
MB2005 x 7
MB2000 x 4
MB2005 x 4
MB2000 x 5
MB2000 x 4
MB2000 x 4
MB2005 x 7
MB2000 x 4
KNMI x 13
MB2005 x 4
KNMI x 6
MB2000 x 8
KNMI x 6
MB2005s x 4
 The 2011 experiments aimed to take advantage of the typically strong eastward tropospheric and stratospheric winds at this location and date; therefore, arrays were deployed primarily to the east of the source (Figure 1b). Table 3 lists the array locations as well as azimuth and distance to the explosion source. IMS arrays to the north and east are IS31 (Kazakhstan), IS43 (Russia), IS46 (Russia), and IS34 (Mongolia). The aforementioned IMS arrays to the west and northwest of Sayarim (IS26 and IS48) also recorded data during the period. NCPA deployed three arrays in northern Israel (IN1, IN2, and IN3) with assistance from the Geophysical Institute of Israel (GII) and Soreq and two lines of single-sensors near the source running ~40 km east and south of the explosion site (with the assistance of GII, UAF, and U.S. Army Research Lab (ARL)). The CTBTO deployed one array in Djibouti (DJ) with assistance from Centre d'Etudes et de Recherche de Djibout and two in Oman (OM_N and OM_S) with Sultan Qaboos University. Joint teams from the NCPA and CTBTO deployed one array in Kuwait (KU) and Qatar (QA) and four arrays in Jordan (JO_NE, JO_NW, JO_S, JO_M) with assistance from the Kuwait Institute for Scientific Research, the National Committee for the Prohibition of Weapons in Qatar, ARL, the Japan Weather Association (JWA), and the Natural Resources Authority in Jordan. UAF and Ilia State University of Georgia installed an array in southern Georgia (GE). ARL deployed an array in Iraq. Note that some sites from the 2009 experiment were also reoccupied due to the potential for westward stratospheric winds in the winter, as well as for comparison between easterly and westerly arrays. The reoccupied sites include: Cyprus (CYPR) by NCPA and Greece (RHOD and PELO) by the CTBTO and NCPA. The majority of the temporary arrays in 2011 (14/17) used the newly designed and constructed NCPA digital infrasound microphones [Alberts et al., 2013]. These sensors have a flat frequency response between ~0.01 and 400 Hz and low noise level, and the vast majority performed well during the experiment. The remaining sensors deployed all have flat frequency responses in the band of interest (~0.1–20 Hz). Noise reduction techniques varied between array locations and experiments, but careful site selection in low wind and forested areas was a priority for all installations. Porous hose installation was standardized for the 2011 experiments to reduce noise levels. The 2011 temporary deployments were coordinated by a team from the NCPA and CTBTO.
Table 3. List of Stations for 2011 Experiment as a Function of Distance From the Source
NCPA IS Mic x 1
NCPA IS Mic x 1
NCPA IS Mic x 1
NCPA IS Mic x 1
NCPA IS Mic x 1
NCPA IS Mic x 4
NCPA IS Mic x 4
NCPA IS Mic x 4
NCPA IS Mic x 5
NCPA IS Mic x 5
MB2000 x 5
NCPA IS Mic x 6
NCPA IS Mic x 4
NCPA IS Mic x 3
NCPA IS Mic x 4
NCPA IS Mic x 1
NCPA IS Mic x 4
NCPA IS Mic x 4
NCPA IS Mic x 3
NCPA IS Mic x 4
MB2005 x 5
MB2005 x 4
MB2000 x 4
MB2000 x 4
MB2000 x 8
MB2000 x 4
MB2000 x 8
3 Data Processing and Methods
 To identify infrasound signals from the Sayarim explosions, array data are processed in the following manner. Data are first band-pass filtered and split into discrete windows. Due to the expected variation in infrasound arrival characteristics at the arrays, variable frequency bands were chosen to capture the peak signal-to-noise ratio (S/N). The frequency band 0.3–5 Hz is chosen for the majority of the arrays in 2011, except for the Jordan and Israel arrays where 0.05–5 Hz is selected. For the 2009 array data, 0.5–5 Hz is chosen to reduce interference from the microbarom peak [Bowman et al., 2005]. The peak explosion periods are about 1–2 s so that most signal phases should have significant energy in the 0.5–5 Hz band. Window durations were typically 10 s, except for JO_S (Jordan) which was 5 s. After filtering, the trace velocity (component of the signal velocity in the plane of the array) and azimuth (apparent geographic bearing from the array to the source) are found for each time window using a least squares solution for plane waves traversing the array [Szuberla and Olson, 2004]. S/N characterization is done using the Fisher Statistic [Melton and Bailey, 1957], a common infrasonic detection technique which performs a comparison of the signal and uncorrelated noise variances to estimate the signal-to-noise power ratio (PS/N) of correlated signals across an array [Olson and Szuberla, 2008]. PS/N is estimated by:
where F is the Fisher statistic and n is the number of sensors in the array. We apply 75% overlap between data segments [Blandford, 1974] and further restrict signals of interest to data segments originating from ±10° of the theoretical azimuth to the explosion location and have an acoustic trace velocity (0.25–0.45 km/s). In this manuscript, we focus on the general identification and characterization of likely signals from the calibration experiments; therefore, we do not set a specific PS/N threshold to identify “detections.” We do note that the majority of signals have high Fisher ratios. Furthermore, other detection techniques (e.g., PMCC) [Cansi, 1995] could be used and should produce similar results. To increase the S/N of the array data, delay and sum beamforming [Johnson and Dudgeon, 1992] is performed.
 Propagation modeling in this study is performed using two different methods. The first is a high-Mach number, planar approximation parabolic equation (PE) method [Lingevitch et al., 2002]. Here we use the PE to predict the transmission loss as a function of range and height. Range-dependent PE simulations are run for each experiment at 0.5 Hz out to 3500 km for 26 August 2009 and 24 January 2011 and 7000 km for 26 January 2011. Absorption is accounted for using estimates from Sutherland and Bass . This attenuation model does well in the troposphere and stratosphere, however, thermospheric absorption is greatly overestimated by the Sutherland and Bass  model compared to observations [Fee et al., 2010; Norris et al., 2010]. Attenuation in the thermosphere is thus currently not well-understood, so we will use the Sutherland and Bass  model with the caveat that thermospheric arrivals are not captured by the model. The model also assumes a rigid ground boundary condition and no topography. Ray tracing is also used to predict arrival times and visualize propagation paths. The ray-tracing method here follows from a range-dependent planar approximation of the full classical Hamilton ray tracing equations (3-D spherical coordinates) found in Gossard and Hooke . Propagating rays that best connect the source and receiver to within a specified tolerance (2 km), termed eigenrays, are selected.
4 Seasonal Winds and Atmospheric Specifications
 Long-range sound propagation is primarily determined by the horizontal wind and temperature gradients in the atmosphere [Drob et al., 2003]. Sound from a surface source primarily radiates upward from the ground due to the generally decreasing sound speed with height. In a stationary atmosphere, upward propagating sound refracts back down to the ground when the static sound speed (c) at altitude exceeds that at the source, creating a waveguide or duct. However, winds in the atmosphere also influence sound propagation. The effective sound speed approximation, valid for shallow angle propagation, gives a qualitative representation at all angles. It is assumed that the propagation is in-plane and that the horizontal winds are small compared to the adiabatic sound speed (Mach number). The effective sound speed ceff is given by the static sound speed plus the horizontal wind component in the direction of propagation:
where γ is specific heat ratio, R is universal gas constant, T is temperature, v is horizontal wind vector, and n is horizontal projection of the ray normal. Thus, to a first order, a sound duct in a moving atmosphere is created where ceff at altitude exceeds that at the source, and sound is preferentially guided downwind. Quantitative propagation modeling using this approximation leads to slightly underpredicted travel times and shadow zone locations, effects that are magnified when the Mach number increases (stronger winds) [Assink et al., 2011].
 Seasonal stratospheric wind jets focused at roughly 55 (winter hemisphere) and 70 km (summer hemisphere) height have been shown to strongly influence the propagation of infrasound [e.g., Donn and Rind, 1971; Drob et al., 2003; de Groot-Hedlin et al., 2010] and hence the number and location of IMS infrasound array detections [e.g., Le Pichon et al., 2009]. In spring and summer, midlatitudes have a relatively stable stratospheric wind jet flowing west. Fall and winter are characterized by a reversal of this wind jet, as it flows predominantly to the east. During this period, the jet is often stronger in magnitude and slightly lower in altitude than during spring and summer but is less stable. Global-scale disruption of the Northern Hemispheric wintertime stratospheric wind jet can occur as a result of sudden stratospheric warming (SSW) events [e.g., Charlton and Polvani, 2007]. Westward stratospheric infrasound ducting in the Northern Hemisphere is possible during winter times as the result of westward winds associated with the polar stratospheric vortex splitting. The cause of SSW events is likely related to planetary waves propagating vertically from the troposphere into the stratosphere [Holton, 2004]. Evers and Siegmund  showed how a major SSW event clearly affected detections at multiple mid- to high-latitude IMS infrasound arrays. In addition to the stratospheric winds, at these latitudes a tropospheric wind jet centered at ~10 km (the jet stream) flows east during spring and summer and generally increases in magnitude and becomes more stable during fall and winter. The consequences of these dominant wind jets are that long-range sound at midlatitudes is preferentially guided in a stable stratospheric duct to the west during spring and summer and in a stronger but less consistent stratospheric duct to the east during fall and winter. If the jet stream is strong enough to overcome the negative lapse rate, sound is ducted to the east in the troposphere as well. In addition to the circulation pattern variability mentioned above, solar heating causes diurnal, semidiurnal, and terdiurnal migrating solar tides in the upper atmosphere that affect refraction in the stratosphere [Donn and Rind, 1972; Green et al., 2012] and thermosphere [Garces et al., 2002; Assink et al., 2012].
 There are several operational atmospheric data assimilation specifications that can be used to compute the instantaneous (or historical) characteristics of infrasound propagation. These specifications (e.g., NOAA GFS [Kalnay et al., 1990]; NASA GEOS-5/MERRA [Rienecker et al., 2008]) are derived from available global satellite- and ground-based observations and are limited to altitudes below 45 and 75 km, respectively. In order to account for thermospheric modes in the infrasound propagation modeling, these available lower and middle atmospheric specifications are typically extended with the NRLMSISE-00 [Picone et al., 2002] and HWM07 [Drob et al., 2008] empirical climatological models using the approach described in Drob et al. . The resulting Ground-to-Space (G2S) atmospheric specifications provide global estimates of the atmospheric state variables, winds, and temperature up to 170 km altitude at 6 h intervals that are specifically tailored for infrasound propagation calculations. Both the European Centre for Medium-Range Weather Forecasts (ECMWF) [Molteni et al., 1996] and G2S specifications based on the NOAA-GFS and NASA GEOS5 have been used extensively in long-range infrasound propagation studies. Neither the G2S nor ECMWF specifications can explicitly resolve small-scale variability (e.g., subgrid-scale gravity waves) in the atmosphere and both must still rely on climatological models above ~70 km. Both small-scale variability and details of the upper atmospheric state can affect infrasonic propagation [Kulichkov, 2010].
 Figure 2 focuses on the zonal wind velocities (east-west, positive east) above Sayarim (~30.0°N, 34.8°E) between 2003 and 2011, taken from the 4 times daily G2S models for this period. Figure 2a is the zonal winds at 12 km (black) and 50 km (red) height. The aforementioned zonal wind magnitude and direction trends are clearly seen. Note that the peak zonal wind velocities occur during the winter, but that the direction is highly unstable during this period. Between the months of November to February 2003–2011, the zonal winds at 50 km blow east 72% of the time. For the period May to August 2003–2011, winds at 50 km are going west 99.2% of the time. Zonal wind velocities between 0 and 100 km for the same period and location are shown in Figure 2b, yellow to red colors indicating eastward propagating wind and blue colors westward. Again, the temporal changes are clearly seen. Note however that the zonal wind variability during the winter in the stratosphere jet peaking near 55 km is not reflected in the westward summer time zonal middle atmospheric wind jet that forms near ~65–75 km and is inherently more stable. Figures 2c and 2d display the zonal winds for the three weeks surrounding the 2011 Sayarim experiments. A SSW occurs prior to the 2011 explosions, causing a change in the stratospheric winds and temperature. If the experiments would have taken place during this period, sound could have propagated to the west being ducted at ~40–50 km and to the east being ducted at ~70 km height. The jet stream magnitudes are also noteworthy, sometimes reaching 70 m/s. The meridional winds (north-south, positive north) are much smaller in magnitude and do not affect stratospheric propagation as significantly as the zonal winds.
 The G2S, denoted by solid lines, and ECMWF, denoted by dotted lines, derived zonal and meridional winds as a function of height above the explosion source for the three Sayarim calibration experiments are shown in Figures 3a–3c. The static and effective sound speeds are also shown (Figured 3d–3f). The ceff profiles are displayed for sound directed 310° for the 2009 experiment and 90° for the 2011 experiments, corresponding to the general direction of deployed arrays (Figure 1). The black line represents ceff0 the effective sound speed at the source, and ducting heights are located where ceff exceeds ceff0. During the 2009 experiment, the winds are characterized by a moderately strong westward zonal wind jet, peaking at ~44 m/s at 62 km altitude (Figure 3a). Meridional winds are mostly minor until the thermosphere. Tropospheric winds have a positive component in both the north and east directions, indicating a weak tropospheric jet to the northeast (~13 m/s peak). For propagation to the west, sound is predicted to be ducted in the stratosphere beginning at ~40 km height (Figure 3b). The wind specifications during the 2011 experiments are dominated by two main features: (1) a strong eastward stratospheric jet, peaking at ~86 m/s at ~64 km height on 24 January and ~79 m/s at ~63 km on 26 January and (2) a strong eastward tropospheric jet (the jet stream), peaking at ~50 m/s at ~14 km on 24 January and ~70 m/s on 26 January (Figures 3b and 3c). Meridional winds are mostly weak, with a slight northerly component. The ceff profile for propagation to the east largely reflects these two wind jets, with significant stratospheric waveguides predicted for both experiments beginning at ~40 km. Furthermore, strong tropospheric waveguides are predicted on both 24 and 26 January (Figures 3e and 3f). There is general agreement between the G2S and ECMWF wind and sound speed profiles below ~40 km. However, significant differences exist above 40 km, particularly in the wind profiles on 26 August 2009. These differences are significant enough to have qualitative impact on the effective sound speed profiles. The detailed propagation modeling and data analyses required to determine whether these differences can be resolved in the acoustic data collected in these experiments will not be undertaken here but will be considered in future work. The GII launched radiosondes near the explosion site and measured atmospheric variables up to 35 km. These data are broadly consistent with the G2S specifications, with the major difference being the slight underestimation of the strength of the jet stream.
 The atmosphere (and thus infrasound propagation) also changes as a function of range. To examine the range dependence, we calculate the effective sound speed ratio (ceff_ratio), defined as the effective sound speed as a function of range and altitude divided by the sound speed at the source: ceff_ratio = ceff(r,z)/c(0,0). Sound ducting is predicted where ceff_ratio > 1. Figure 4 shows the effective sound speed ratio for the three Sayarim experiments out to 3000 km for azimuths of 310°, 60°, and 60° respectively for the 2009 and two 2011 explosions. Red colors indicate ducting conditions. For the 2009 experiment, the two stratospheric turning heights are present at ~45 and 65 km, with the 65 km duct increasing initially and then both weakening after ~1500 km, indicating a marginal stratospheric duct at this range. Range dependence also exists in 2011, and stratospheric ducting is much stronger. The ceff_ratio in the stratosphere is well above 1.0 on both 24 and 26 January between ~40 and 75 km. Although range dependence for the stratospheric duct exists, it remains well established out past 3000 km for these profiles. Changes in the jet stream clearly affect the ceff_ratio for both experiments and weakens the duct after ~1000 km. Although the ceff_ratio for north-south propagation is not shown, it also contains significant range-dependent variations. The range dependence of the lower atmosphere over the ranges of interest are driven by land-sea contrasts, topography, and other localized sources of vertical convection; in contrast to the stratosphere and upper atmosphere where the in situ dynamical forcing terms are much more homogeneous and globally connected, with the exception of some possible range dependence created by buoyancy waves that propagate upward from the troposphere [Holton, 2004].
5 Explosion Characterization
 Charge design and TNT yield estimations are presented in detail in Gitterman and Hofstetter . All three explosions were successfully conducted by the GII on the ground surface, at the Sayarim Military Range in southern Israel (Figure 1). The GII recorded all explosions with near-source high-pressure gauges (in the 200–600 m distance range), accelerometers, and high-speed video cameras.
 The first Sayarim experiment consisted of ~82 t of cast chemical explosives (higher explosive energy density than TNT) placed in large barrels and assembled in a compact pyramidal shape. Successful detonation of the explosives occurred on 26 August 2009 06:31:54 UTC at 30.000 57°N, 34.81351°E [Gitterman, 2010]. The majority of the explosion energy was directed upward by the preferential placement of mines in the booster to focus the energy into the air rather than the ground (the same upward detonation concept was applied for the 2011 shots). High-pressure gauges recorded the airblast waves, and explosive yield estimation based on these data give an equivalent TNT yield of ~96 t, an ~17% increase over the actual weight [Gitterman, 2010]. High-speed video also recorded the explosion.
 The January 2011 explosions were conducted at a nearby location on the Sayarim Military Range and consisted of two separate explosions, using bulk ANFO explosives in big bags and assembled in a roughly hemispherical shape. The first was a 10.24 t shot detonated on 24 January 2011 13:17:54 UTC at 29.99555°N, 34.81668°E [Gitterman et al., 2011]. The larger explosion consisted of 102.1 t of explosives detonated on 26 January 2011 07:17:43 UTC at 30.00064°N, 34.81324°E. The 24 January explosion was conducted as a “test” explosion and to provide further comparison between events, in particular one with a comparable but different atmosphere. Explosion yields were determined from the high-pressure gauges in a similar manner as the 2009 experiment and are estimated to be 7.4 and 76.8 t of TNT equivalent [Gitterman and Hofstetter, 2012], lower than the charge weights. Gitterman et al.  suggest that the lower yields resulted from energy loss from nonhemispherical and nonhomogenous charges and air voids between the charge units. Peak periods for the explosion signals are between ~1 and 2 s. Figure 5 shows the pressure signal recorded at 5.6 km from the 26 January explosion (a) as well as a photo of the explosion and association shock wave (b). More detailed analysis and discussion of the near-field pressure data and explosion source are discussed in Bonner et al.  and Gitterman and Hofstetter 
 The signals from the explosions recorded at each array are shown in Figures 6-8 for 2009 and Figures 10 and 11 for 2011. Each figure subplot shows the received waveform, travel times, celerities (defined as range to the source divided by travel time), trace velocities, and back azimuths along the waveform. The frequency bands, time windows, and step sizes used for these computations are listed in section 3 and were chosen specifically for the different arrivals to maximize the visible energy. The celerities and absolute travel times of the different portions of the wave train are indicated on the axis below the waveform. Trace velocities and observed back azimuths along the wave train have been determined and are plotted below the wave train in gray scale. The gray scale indicates the Fisher ratio for the corresponding time window and is a measure of the statistical significance of the detection. The solid horizontal lines correspond to the sound speed on the ground and to the actual back azimuths.
 Celerities, trace velocities, and apparent back azimuths can be used to infer properties of a signal phase and its propagation path. The celerity is a measure of the horizontal propagation speed and reflects both signal propagation speed and path length. As a consequence, the closer the propagation path is to horizontal, the closer the celerity is to the ground sound speed. The various phases of an infrasonic wavetrain recorded by an array on the ground tend to cluster into groups determined by the type of propagation path. Signals traveling along tropospheric paths tend to arrive first, with celerities closer to the ground sound speed, followed by stratospheric arrivals with smaller celerities and then by thermospheric arrivals. Typical stratospheric celerities range between 0.28 and 0.31 km/s while typical thermospheric celerities are 0.23–0.28 km/s [e.g., Brown et al., 2002]. The trace velocity is equal to the cosine of the grazing angle times the sound speed at the ground and thus the closer the trace velocity is to the ground sound speed, the closer the signal grazing angle is to 0°. In addition, the trace velocity is equal to the effective sound speed at the altitude at which the propagation path turns back toward the ground. Signal phases with trace velocities significantly higher than the ground sound speed are either from an elevated source or have returned to the ground along a path which has ascended to an altitude at which the (effective) sound speed is approximately equal to the trace velocity. Finally, the bearing deviation, defined as the difference between the apparent back azimuth and the actual back azimuth to the source, is a measure of the cross winds encountered by the signal along its propagation path [Diamond, 1963]. In identifying signal phases and their propagation paths the bearing deviation can be used as a consistency check. To produce a significant bearing deviation, the signal must pass through a wind jet flowing perpendicular to the propagation direction.
6.1 2009 Infrasound Array Detections
 Infrasound from the 2009 explosion is detected at 13 arrays to the west and north of the detonation site, out to 3428 km (CEA array). Figure 1a shows a map of the arrays, with filled markers representing arrays that detected the explosion using the detection methods outlined in section 3. Individual array detection are shown in Figures 6 (out to 2000 km) and 7 (>2000 km). Beamformed array data as a function of distance from the source out to 4000 km are shown in Figure 8. Array IMA (190° azimuth to explosion, 339 km distance from the source) has two clear sets of arrivals, with the first being impulsive and highest amplitude. The detections here likely correspond to tropospheric and stratospheric arrivals, based on the observed celerities (Figure 6). Arrays RHOD, PELO, and CALA (115°–135°, 941–1945 km) all show similar characteristics (Figure 6). Detections at these arrays consist of two sets of arrivals: the first being higher amplitude with celerities between 0.33 and 0.27 km/s and the second of lower amplitude with celerities between ~0.24 and 0.20 km/s. CYPR appears to record both sets of arrivals, but due to its configuration (line array), we are not able to resolve the azimuth, hence its resolution is fixed at an azimuth of 162° and we solve for trace velocity only. Processing in a more defined band (~1-5 Hz) at CRET also shows evidence of later arrivals. Figure 9 details the signals from the low-noise PELO array located ~1375 km NW of Sayarim. The aforementioned two sets of arrivals are clear. The first consists of multiple pulses of energy between ~07:41:30 and 07:52:00 UTC, corresponding to celerities between ~0.33 and 0.29 km/s. The second set of arrivals are lower amplitude between ~08:06:00 and 08:24:30 UTC, with celerities ranging from 0.24 to 0.20 km/s (Figures 9a and 9b). Also of note is the frequency content of the arrivals. The spectra (Figure 9c) shows high-frequency energy (up to 6 Hz) above the background noise for both sets of arrivals, although the first set of arrivals is centered at lower frequency (~0.3 Hz compared to ~1–3 Hz for the second set). Beyond 1943 km, only a single set of arrivals are detected at stations IS48, IS62, IS63, IS26, PROV, IGA, and CEA (Figures 7 and 8). Signal durations generally increase as a function of range until ~2750 km, where higher noise levels (and thus lower S/N) begin to reduce observed durations. The IMS arrays IS26 and IS48 clearly detect the explosion as well. No detections are made beyond the CEA array (3428 km) at FLERS, DBN, DIA, TEX, or IMS arrays to the south or east of Sayarim.
6.2 2011 Infrasound Array Detections
 The 2011 explosions are clearly detected on multiple arrays to the east and north of Sayarim, with the most distant station being IS34 (~6300 km) (Figure 1b). Figures 10-13 show the waveforms and array processing results from regional and remote stations that detected the 24 and 26 January explosions. The plots are grouped, roughly, by range and geographic region. The middle and northern arrays in Jordan (JO_M, JO_NW, JO_NE) lie roughly on the same azimuth and are grouped together. Similarly, the arrays in northern Israel (IN1, IN2, IN3) are roughly at the same azimuth and are grouped together. The array in southern Jordan (JO_S) is an outlier, both in range (at 104 km, it was squarely in the classical shadow zone) and azimuth. The Kuwait (KU), Qatar (QA), and Georgia (GE) arrays are grouped together by virtue of being at ranges intermediate between the regional network and the IMS stations. Finally, the IMS arrays in Kazhakstan, Russia, and Mongolia (IS31, IS46, IS34) are also grouped together. The northern Oman array (OM_N) detected both the 24 and 26 January explosions with low signal noise while the southern Oman array (OM_S) did not detect either, partially due to high noise levels.
 The 24 January explosion is clearly recorded by the regional arrays to the north and east of Sayarim in Israel and Jordan (Figures 10 and 11), extending to OM_N at 2420 km. No detections are made at any IMS stations or any of the stations to the west. Based on celerity and trace velocity, all of the Jordan arrays (JO_S, JO_M, JO_NW, JO_NE) and the Kuwait array (KU) appear to have detectible and, in most cases strong, tropospheric arrivals. With the exception of JO_S, which was well within the “classical stratospheric shadow zone” [e.g., Gutenberg, 1926], all these arrays appear to record stratospheric arrivals as well. The signal detected in Qatar (QA) was not very strong, but stratospheric returns appear to have been detected, and possibly tropospheric as well, but with little S/N. Stratospheric returns and a single thermospheric return were detected on the two arrays in northern Israel (IN1, IN2).
 Detection of the 26 January explosion was widespread to the east and north of Sayarim (Figures 1b, 12, and 13). Signals from the event were detected at all the arrays deployed in northern Israel and Jordan; at the arrays in Qatar, Kuwait, and Georgia; and at three IMS stations, IS31 in Kazakstan, IS46 in Russia and IS34 in Mongolia.
 Conditions had changed between 24 and 26 January (Figures 2c and 2d). By the time of the 76.8 t explosion on 26 January, a low altitude (about 800 m) wind jet flowing toward the southeast had developed. This wind jet was of great significance for the near-field [Bonner et al., 2012]. At higher altitudes, the jet stream had increased to about 75 m/s, and the stratospheric jet had turned slightly to the north. In addition, there were rainstorms along the Persian Gulf leading to high noise levels at the stations in Kuwait, Qatar, and Oman.
 All three arrays in northern Israel (IN1, IN2, IN3) show what appear to be stratospheric arrivals with celerities ranging from 0.27 to 0.31 km/s (Figure 12a). There appears to be a thermospheric arrival detected on all three arrays with celerities between 0.24 and 0.26 km/s. A possible second thermospheric arrival is seen on IN3 with celerity of about 0.225 km/s. We do not make any claims on the U-shaped signal observed at celerities of about 0.286, 0.292, and 0.295 km/s on IN1, IN2, and IN3, respectively. Strong tropospheric arrivals, characterized by celerities of about 0.32–0.33 km/s and trace velocities close to 0.34 km/s, are seen on the three arrays along the northeast line in Jordan: JO_M, JO_NW, JO_NE (Figure 12b). On JO_NW and JO_NE somewhat weaker stratospheric arrivals follow the tropospheric phases with celerities ranging from 0.29 to 0.31 km/s and trace velocities of 0.36 km/s and higher. A possible thermospheric arrival is visible in the record from JO_NW. A very large tropospheric signal was recorded on the JO_S array (Figure 12c).
 As pointed out above, noise levels along the Persian Gulf were high on 26 January. A weak detection was made in northern Oman, and those signals detected in Qatar and Kuwait were not much larger than the background atmospheric noise (Figure 13a). Despite this, what appear to be tropospheric arrivals were detected on both QA and KU, despite the ranges being well in excess of 1000 km. Stratospheric arrivals appear to have been detected at QA as well. Similarly, in Georgia, at GE, signals with both stratospheric and tropospheric celerities are observed over the atmospheric noise. The signals detected at the three IMS stations, IS31, IS46 and IS34, consist of extended waveforms typical of very long range detections and appear to be stratospheric in origin (Figure 13b).
 As expected, the 2009 experiment showed enhanced propagation to the west of the explosion source due to the westward stratospheric wind jet. However, the full azimuthal dependence of the propagation is not well resolved as the vast majority of stations were deployed to the northwest of the explosion source. Most of the array detections have celerities corresponding to stratospheric arrivals, and multiple energy pulses at more distant stations likely correspond to multiple stratospheric arrivals. Stratospheric arrival durations generally increase with distance, as expected from multipathing (multiple stratospheric arrivals reaching the ground). Eventually, decreased amplitudes from greater attenuation and higher noise levels reduce the observed durations. The eastward flowing stratospheric jet during the 2011 experiment is significantly stronger than the westward flowing jet during the 2009 experiment. In addition, on both 24 and 26 January, there was a significant eastward flowing jet stream at about 15 km altitude, reaching about 50 m/s on 24 January and about 70 m/s on 26 January. The jet stream produced large tropospheric arrivals to the east while the stratospheric jet enabled detections at very long range. A tropospheric jet blowing to the northeast also exists for the 2009 experiment, but the lack of infrasound arrays in this region leaves the infrasound propagation unconstrained.
7.1 26 August 2009 Propagation
 Ray tracing and PE modeling to the PELO array are shown in Figure 14a. Predicted stratospheric celerities for the PELO array from ray tracing are ~0.304 km/s (~4530 s travel time), in decent agreement with the observed celerities of the first arrivals (0.33–0.29 km/s), particularly the highest amplitude arrivals at PELO. Only eigenrays being refracted around 60 km are predicted to arrive at PELO. However, the gradient of the ceff in the stratosphere during this time period is relatively weak (Figure 3d), which would result in sound energy being refracted over a range of heights and therefore a broad range of celerities. This is apparent in the PE modeling in Figure 14a. We also note the discrepancies between the G2S and ECMWF profiles at these heights (Figures 3a and 3d). Small-scale variability not captured by the atmospheric specifications may also contribute [Green et al., 2011]. The second set of 2009 observed arrivals at PELO have celerities (0.24–0.20 km/s) on the low end of that expected for thermospheric arrivals. Ray tracing predicts thermospheric arrivals (Figure 14a) with celerities ~0.235 km/s (~5855 s travel time), consistent with the fastest portion of the second set of arrivals. Low-celerity (<0.23 km/s celerity) arrivals are not predicted by the ray tracing or PE modeling, and thermospheric arrival amplitudes are predicted to be very low (Figure 14a). Although the second set of arrivals have celerities similar to typical thermospheric arrivals, the frequency content is unexpected. Nonlinear propagation effects and acoustic attenuation are both very strong in the thermosphere. Typically, thermospheric arrivals have longer periods and lack energy at higher frequencies due to nonlinear propagation effects and severe absorption, compared to stratospheric arrivals [Rogers and Gardner, 1980; Kulichkov, 2002]. The second set of arrivals have higher-frequency content compared to the first set (Figure 9c). Although thermospheric arrivals are not unexpected, the relatively low celerities (down to ~0.19 km/s), relatively high amplitudes, and high-frequency content are not predicted by the propagation model results. These arrivals are thus likely due to features not resolved by the atmospheric specifications [Chunchuzov et al., 2011; Drob et al., 2013]. For example, sharp wind jets in the mesosphere have been proposed as a potential reflector of sound energy [Kulichkov, 2010] and would effectively high-pass filter the sound energy. A sharp mesospheric wind jet may therefore be responsible for the second set of arrivals in 2009 and helps explain the abundant high-frequency energy and long travel times. The travel times (and low celerities) are likely due to longer path lengths in low velocity zones such as the mesosphere. During this period the ceff in the mesosphere along this path is quite low (down to 210 m/s at 92 km, Figure 3d), indicating slow propagation velocities. Mesospheric wind jets at ~90 km would not be well constrained by the G2S models due to their use of climatological versus measurement-based models at this altitude. Other studies have reported mesospheric arrivals not predicted by propagation calculations in conjunction with available atmosphere specifications [Kulichkov, 2010; Assink et al., 2012]. Higher than predicted amplitudes from upper atmosphere arrivals have also been documented elsewhere [Fee et al., 2010; Norris et al., 2010], suggesting a potential overestimation of absorption in the mesosphere and thermosphere by Sutherland and Bass . Lastly, we note that although the two sets of arrivals have different frequency content and celerities, their qualitatively similar amplitude envelopes in the filtered waveforms (Figure 9b) is likely a consequence of some similarities in their propagation paths.
7.2 January 2011 Propagation
 Acoustic propagation for the 2011 explosions was better than expected, with infrasound being recorded out to nearly 6300 km. Although the yield of the 26 January 2011 explosion was less than the 26 August 2009 explosions (76.8 versus 96.0 t TNT equivalent), the stronger stratospheric wind jets and deeper sound ducts during this time permitted longer range detection. Clear tropospheric and stratospheric arrivals are observed on numerous stations, and complexity in the arrival structures exists due in part to the multiple wind jets and range dependence. The closely spaced Israel arrays to the north have a qualitatively similar, yet complicated arrival structure (Figures 10a and 12a) with multiple arrivals likely corresponding to stratospheric and thermospheric phases (based primarily on the celerities, trace velocities, and examination of the atmospheric profiles). These arrays also have long-duration codas after the main arrivals, possibly due to scattering from atmospheric inhomogeneities. The Jordan arrays along a line to the northeast have a complicated arrival structure as well. Multiple tropospheric and stratospheric arrivals are observed, with the tropospheric jet filling the classic “shadow zone” between ~50 and 250 km. The arrival structure at these arrays suggests complex sound propagation in both the tropospheric and stratospheric ducts. Figure 14b shows the ray tracing and 0.5 Hz PE modeling for propagation to the northeast on 26 January. Strong tropospheric and stratospheric ducting is predicted, consistent with the numerous tropospheric and stratospheric arrivals observed. Weak thermospheric ducting is also predicted and observed. Note the lower TL predicted along the ground for the 2011 experiment versus 2009, consistent with the greater detection range. Due to the large number of arrays and variability in propagation and infrasound arrivals, detailed arrival identification and analysis for the 2011 experiments will be examined further in other studies.
 Upper atmospheric arrivals are also not as prevalent in 2011 compared to 2009. The deep waveguides in 2011 will duct a higher fraction of sound energy in the troposphere and stratosphere (compared to 2009), decreasing the amount of energy available to refract in the thermosphere (Figure 14). Unlike the 2009 experiment, arrays were also deployed opposite the direction of the stratospheric wind jet for the 2011 explosions. These westerly stations did not detect the explosions, again verifying the strongly anisotropic propagation. Note that upper atmospheric returns were not detected at these westerly stations, despite their unambiguous detection during the 2009 experiment.
 It is clear from this study and previous work that for surface explosions with yields >75 t at nonequatorial latitudes during the solstice periods, detection is likely at multiple IMS stations within ~3500 km. During periods of strong stratospheric ducting (typically winter), detection by numerous IMS arrays out to >6000 km is possible, but of course dependent on local wind noise conditions. However, as noted by Le Pichon et al.  and Green and Bowers , due to the anisotropic nature of long-range infrasound propagation, IMS array detections will primarily occur in an east-west line near the solstices and have poor azimuthal resolution, making accurate source locations difficult. Also, due to the strong winds the horizontal translation of the sound energy (and thus observed azimuthal deviation) will be large, increasing modeling and source location errors. We note that integration of seismic data [Mialle et al., 2011] may help constrain the source location.
7.3 Predicted Transmission Loss
 The range-dependent transmission loss (TL) at 0.5 Hz was calculated as a function of range and azimuth for all three explosions out to 3500, 3500, and 7000 km for the respective experiments (Figure 15). This figure clearly illustrates the effect of various waveguides and corresponding wind jets on propagation, with warmer colors indicating low TL (higher observed amplitudes). As expected, sound is predicted to be broadly guided to the west during the 2009 experiment due to the stratospheric wind jet, with a northerly propagation component also apparent. Multiple regions of relatively low TL within the first ~1000 km west of Sayarim correspond to stratospheric “bounce points” where stratospheric energy concentrates at the ground at down-wind intervals approximately every 250 km. Of note is the relative strength of the tropospheric ducting predicted to the northeast due to the jet stream. The duct shows clear range dependence and becomes much weaker after ~1100 km. The low TL predicted for this duct is due to the relatively low amounts of absorption and shorter propagation paths in the troposphere. Very high attenuation is predicted to the south, with only thermospheric arrivals (and thus high absorption) predicted.
 TL estimates for the 2011 experiments differ from the 2009 experiment (Figures 15b and 15c). Propagation to the east is generally very efficient. Both 2011 experiments have strong, broad waveguides to the east, southeast, and northeast due to the strong tropospheric and stratospheric winds. Propagation to the east for both of these experiments is predicted to be stronger than propagation to the west during the 2009 experiment due to the stronger and deeper zonal wind jets, which is consistent with the greater detection distance. Southeast propagation is predicted to be stronger on 24 January versus 26 January. Severe TL is predicted to the south and west for these experiments due to thermospheric attenuation. The lack of acoustic shadow zones within 250 km to the east for these events is due to the tropospheric ducting, instead the region is much more ensonified.
 Array detections qualitatively match the predicted transmission loss for all three experiments as shown in Figure 15. Arrays that detected the explosion are denoted by a filled symbol, while a nondetection is represented by open symbols. The IMA array to the north-northeast of Sayarim detected arrivals from the 2009 experiment, as predicted by the TL modeling. The IS31 array far northeast of Sayarim did not detect the explosion, consistent with the modeling. Relatively small TL (<120 dB) is predicted to most arrays to the northwest (e.g., RHOD, PELO), consistent with the fact that arrays out to 3500 km detected the explosion. Of note however is the detection by numerous arrays past 2000 km northwest of Sayarim (e.g., IS26, IS62, IS63, IGADE). The TL is predicted to be relatively high in this region (>150 dB relative to 1 m). Several explanations are possible: first, it is possible that the stratospheric duct has a stronger northerly component than is provided by G2S specifications; second,, scatter (or partial reflection and refraction) from small-scale inhomogeneities may also contribute [Kulichkov, 2010; Matoza et al., 2011; Drob et al., 2013]. The arrays in Netherlands did not detect the explosion, and much higher TL is predicted here due to the slightly more northerly location and greater distance. Unfortunately, the lack of arrays to the south and east of Sayarim within ~3000 km does not allow comparison with the predicted TL. Most arrays to the east and northeast of Sayarim in 2011 detected the explosions, consistent with the modeling. Of note are the Oman arrays, which lie in a region of relatively low TL, yet only OM_N detected the 2011 explosions. The local noise levels were relatively high for these arrays, and OM_S is also on the southern edge of low TL (Figures 15c and 15d). None of the arrays to the west or south detected the 2011 explosions, consistent with the modeling.
7.4 Future Research
 The data set from these calibration experiments holds much potential for future research and benchmark studies. Studies to determine long-range infrasonic source locations and yield estimates can be performed, as the source location, timing, and yield are exceptionally well constrained. Detailed studies on attenuation and the comparison of results with other available atmospheric specifications (e.g., ECMWF), in particular considering the differences in the specifications of the middle and upper atmosphere (e.g., Figures 3a and 3d), should be undertaken. The second set of arrivals in 2009 is particularly noteworthy, as they are unexplained using the current atmospheric specifications. Also of note is the high station density within the first 500 km of the 2011 experiment, which will permit detailed modeling and interpretation of individual tropospheric, stratospheric, and thermospheric arrivals. We also note the vast majority of past calibration experiments and infrasonic propagation studies focused on the stratospheric wind jets at midlatitudes. Detailed studies on unexplained individual arrivals, such as the U-shaped signals with stratospheric celerities recorded at the Israel arrays in 2011 (section 6.2), would also be a worthwhile area of research. Given the varied nature of the meteorology and middle and upper atmospheric dynamics at different latitudes, further research should also be undertaken on equatorial and high-latitude propagation. Additional propagation and detection studies should be undertaken during the equinox periods. Lastly, comprehensive, quantitative work comparing array detection capability against local noise levels (and other phenomena) would also be beneficial.
 Three large-scale infrasound experiments took place in the Middle East, Europe, Africa, and Asia during August 2009 and January 2011 to test the IMS infrasound network and provide ground truth data for infrasound propagation studies. These experiments clearly demonstrate that infrasound can propagate to extremely long ranges in the waveguides created by the atmosphere winds and temperature gradients. The successful planning and detonation of large chemical explosive charges produced infrasound that was detected out to ~3500 km in 2009 and ~6300 km in 2011. For the 2009 experiment, infrasound was detected by 13 infrasound arrays to the west and north, largely due to the westward stratospheric winds. The 2011 experiments occurred during periods of strong tropospheric and stratospheric wind jets. As expected, these wind jets created strong waveguides to the east and north. Even though the yield was only 7.4 t, the 24 January explosion was clearly recorded on numerous regional arrays and out to ~1690 km. The 26 January explosion was widely recorded by 13 arrays to the east and northeast including three IMS arrays out to ~6300 km.
 The arrays recorded a variety of different infrasound arrivals from the calibration experiments. Two separate sets of arrivals were clear on multiple arrays in 2009, with the first corresponding to well-predicted stratospheric arrivals. The second set of arrivals is not predicted by the atmospheric models and has unusually high-frequency content and low celerities. Reflections from unresolved wind jets in the upper atmosphere may be responsible for these arrivals. Arrays within 500 km of the 2011 events recorded multiple tropospheric, stratospheric, and thermospheric arrivals and show complicated waveforms. More distant arrays to the east show single, low s/n arrivals. Tropospheric arrivals were apparent out to ~1690 km. No detections were made in directions opposite to the tropospheric and stratospheric wind jets for any of the explosions, consistent with basic theory.
 Comparison of predicted transmission loss and observed infrasound arrivals gives qualitative agreement. Propagation modeling for the 2011 experiments predicts lower transmission loss in the direction of the downwind propagation compared to the 2009 experiment, consistent with the greater detection distance. The strong wind jets largely control the observed arrivals and further illustrate the importance of zonal wind jets on propagation and detection. Overall, the atmospheric specifications up to 65 km height suffice for qualitative propagation modeling. However, significant attenuation predicted in the upper atmosphere is inconsistent with upper atmospheric infrasound arrival observations. Unresolved atmospheric features are likely to be the cause of additional propagation paths that are not predicted by the atmospheric specifications used. Additionally, observations suggest a more northerly component to the stratospheric winds in 2009 than that present in the atmospheric specifications.
 The 2009 and 2011 Sayarim experiments further illustrate the complexity and variability of the atmosphere and demonstrate the utility of large-scale calibration experiments with dense networks for better understanding infrasound propagation. These experiments provide rich data sets for future research that will aid our understanding of the atmosphere and infrasound propagation. Lastly, they represent successful collaborations between numerous institutions and countries and demonstrate how international cooperation is crucial for successful infrasound experimentation.
 We gratefully acknowledge all participants involved in the planning, deployment, and analysis of the Sayarim calibration experiments and data. These experiments would not have been possible without the help of countless persons from a wide variety of countries and organizations.
 Many organizations and persons participated in the preparation of Sayarim calibration explosions, near-source measurements, and data processing. High-quality explosives in convenient packages were supplied by IMI Ltd. (I. Veksler) for the 2009 experiment and by EMI Ltd. (Dr. D. Hershkovich) for the 2011 experiment. Elita Security Ltd. (S. Kobi) assembled the 2009 charge with maximal concentration of explosives. The IDF Experiment Division (E. Stempler, Y. Hamshidyan) provided appropriate territory, logistics, and near-source measurements and assembled the 2011 ANFO charges with the optimal initiation/detonation scheme. GII personnel helped in logistics procedures, preparation and deployment of numerous near-source local observation systems, and data processing (U. Peled, N. Perelman V. Giller, A. Schwartzburg). Research work of one of the authors (Y.G.) was supported by the Israel Ministry of Immigrant Absorption. The MERRA/GEOS-5 data utilized in the G2S atmospheric specifications were provided by the Global Modeling and Assimilation Office (GMAO) at NASA Goddard Space Flight Center through the online data portal in the NASA Center for Climate Simulation. The NOAA GFS, also utilized in the G2S specifications, was obtained from NOAA's National Operational Model Archive and Distribution System (NOMADS), which is maintained at NOAA's National Climatic Data Center (NCDC). ECWMF specifications were provided by Laslo Evers. This publication was made possible through support provided by U.S. Army Space and Missile Defense Command under Contract No. W9113M-06-C-0029. The opinions expressed herein are those of the authors and do not necessarily reflect the views of the U.S. Army Space and Missile Defense Command. The manuscript was substantially improved by reviews from David Green, Laslo Evers, and an anonymous reviewer. Distribution Statement A: Approved for public release; distribution is unlimited.