1. Introduction
A tsunami is a long ocean surface wave caused most often by an undersea or coastal earthquake. As a tsunami travels across the ocean, its movement generates gravity waves in the atmosphere. Numerical simulations indicate that these tsunami-generated atmospheric gravity waves can propagate rapidly upward and within a few hours reach peak amplitudes at altitudes of 250 km or higher (e.g., Hickey et al. 2009; Occhipinti et al. 2011). An improved understanding of tsunami-generated gravity waves could conceivably have practical applications for early tsunami detection and warning through near real-time atmospheric monitoring (Artru et al. 2005; Lovett 2010; Liu et al. 2006).
Like the tsunami itself, the tsunami-generated gravity waves have long horizontal wavelengths (hundreds of kilometers), fast horizontal phase speeds (~200 m s−1), and small amplitudes at the ocean surface (tens of centimeters). These properties allow the gravity waves to propagate upward through the lower and middle atmosphere without much geometrical spreading or critical-layer interaction, and without significant dissipation or wave breaking, even as the gravity wave amplitudes grow exponentially with height in response to the decreasing air density. One process that can affect tsunami-generated gravity waves in the lower and middle atmosphere is partial reflection, as noted in the early work of Peltier and Hines (1976).
Here we examine the partial reflection of tsunami-generated gravity waves. The altitude range of our model extends from the ocean surface into the lower thermosphere, to a maximum altitude of 180 km. We solve a vertical eigenfunction equation for the gravity waves and find that each vertical eigenfunction has multiple turning points—at least eight in one example. The gravity waves are thus converted between propagating and evanescent waves several times during upward propagation. The amplitude of the transmitted waves is significant and comparable to the amplitude of the reflected waves. The problem is related to wave tunneling through a series of thin evanescent layers (Fröman and Fröman 2002; Brown and Sutherland 2007).
Hickey et al. (2009) and Occhipinti et al. (2006, 2008, 2011) also calculate solutions related to the vertical eigenfunctions for tsunami-generated gravity waves. Their models extend to higher altitudes than ours and include processes relevant to the ionosphere that we omit, such as wave coupling to the plasma. Hickey et al. (2009) solve the Navier–Stokes equations for each Fourier component of the gravity wave spectrum. Occhipinti et al. (2006, 2008, 2011) solve a set of inviscid equations for each Fourier component. In the latter case, the model does not include downgoing reflected waves or background winds.
We supplement these studies with a more detailed analysis of partial reflection and partial transmission for altitudes up to the lower thermosphere. The present model is anelastic, inviscid, steady state in the reference frame moving with the tsunami, and two dimensional in a vertical plane with a height-dependent atmospheric background, including winds.
In one example, we use this model to study atmospheric gravity waves generated by the devastating tsunami that resulted from the Sumatra earthquake on 26 December 2004 (Titov et al. 2005). Gravity waves generated by this tsunami were also modeled by Hickey et al. (2009) and Occhipinti et al. (2006). As in these studies, we look at a location on the equator over the Indian Ocean west of Sumatra. Ionospheric measurements taken from as far away as Puerto Rico have identified wavelike disturbances interpreted as atmospheric gravity waves generated by the Sumatra tsunami (Lee et al. 2008).
The model formulation is given in section 2. The numerical technique for the vertical eigenfunction equation is described in section 3. An accompanying ray-tracing integration is introduced in section 4. The results are presented in section 5 for propagation through three different background atmospheres: a uniform background, a localized wind jet centered at 100-km altitude, and a realistic vertical profile of wind and temperature representative of atmospheric conditions at the time and location of the Sumatra earthquake. A summary of our main findings is given in section 6.
2. Formulation
The model is linear, two dimensional, and steady state in a coordinate system (x, z) that moves horizontally at the phase speed c of the tsunami. This is a standard formulation for this application (e.g., Hickey et al. 2009; Artru et al. 2005). As in Hickey et al. (2009), we set c = −200 m s−1, corresponding to an ocean depth of about 4 km. For simplicity, we assume that the tsunami and gravity waves propagate in the east–west plane, so that x is a zonal coordinate and negative x is westward. The undisturbed ocean surface is at z = 0.
We mention a few characteristics of the tsunami, some of which can be found in Hickey et al. (2009). The wavenumber of maximum spectral amplitude 
The gravity waves are stationary in the tsunami reference frame. The vertical velocity of the gravity waves is (ρ0/ρ)1/2w(x, z), where ρ(z) is the background density with value ρ0 at z = 0. The density factor (ρ0/ρ)1/2 accounts for non-Boussinesq effects in the anelastic approximation [Smith 1979, his (2.22)] and increases exponentially with altitude. The plots in this paper show w(x, z) without the density factor in order to reveal the solution more clearly at all altitudes.
The vertical eigenfunction in (6) and (7) is derived in Smith [1979, see his (2.23)]. An extension to three-dimensional flow with compressibility effects is derived by Simard and Peltier (1982).
3. Numerical solution of the vertical eigenfunction equation
The numerical method for the vertical eigenfunction equation is standard and involves integrating (6) as two coupled first-order ordinary differential equations for 
We compute WKB solutions at altitudes where ϵ < 1. Broutman et al. (2009, p. 486) used and tested this criterion for an application with mountain waves and turning points. A smaller range of ϵ < 0.5 slightly decreases the extent of the region where WKB solutions are calculated (as plotted here in Figs. 5 and 10).
We found in our examples that when (14) is used instead of (13), our main conclusions are the same: there is an approximate standing wave pattern below the turning points indicating significant partial reflection, and there is significant partial transmission through the turning points. The results presented below are obtained using (13) [except for the example of Fig. 7, which is calculated using (14) and is included for comparison].
4. Ray tracing
The initial condition k(x, z = 0) for the ray tracing is obtained from the tsunami profile at the ocean surface. The tsunami is treated as a slowly varying wave train with a locally defined horizontal wavenumber K(x). The wavenumber K(x) is calculated from the vertical velocity W(x) in (2) and (3) using a Hilbert transform (Hahn 1996). The Hilbert transform gives an imaginary part for W(x), which can be added to the real-valued W(x) to form a complex function. The phase ϕ(x) of that complex function is then differentiated (using a finite-difference approximation) to obtain the local wavenumber K(x) = ϕx. The initial condition for the ray tracing is k(x, z = 0) = K(x).
Estimating the local wavenumber in this way is a well-known signal-processing application of the Hilbert transform. The Hilbert transform is numerically implemented here by the standard method involving a one-sided Fourier transform. Further information on the one-sided Fourier transform and the phase calculation from the Hilbert transform can be found in Hahn (1996, his sections 1.7 and 1.13).
Figure 1 shows W(x) in the top panel and the associated wavenumber K(x) in the bottom panel. For the ray tracing, we use the K values between the plotted circles in the bottom panel of Fig. 1. The circles have x values of −180 and 750 km and K/K0 values of −0.46 and −1.6. Over this range |Kx/K2| < 0.2, which is small compared to unity and thus an indication of a slowly varying wave profile and a well-defined local wavenumber K(x). The position at which K(x) = −K0 is x ≃ 360 km.
(top) The vertical velocity W(x) at the ocean surface normalized by its peak value W0. (bottom) The horizontal wavenumber K(x) normalized by the wavenumber K0 of the maximum 
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
As the rays propagate upward, they encounter evanescent regions where m is imaginary and the ray path is not defined. In an evanescent region, we simply continue the ray straight upward (without horizontal displacement) until the waves are just above the evanescent region and become propagating again, with m real. This seems to be adequate for the thin evanescent regions in the results below. We ignore downgoing rays reflected from turning points.
5. Results
We consider three model atmospheres: a uniform background (case 1), an idealized wind jet (case 2), and a background obtained from empirical atmospheric models for the Sumatra tsunami of 26 December 2004 (case 3).
The Fourier integral for w(x, z) in (5) is approximated by an FFT with 1024 points and a grid spacing of 15 km in x. In the following we will sometimes refer to the local complex magnitude of the vertical velocity. We denote this by |w|, a function of x and z that is obtained from w(x, z) using a numerical Hilbert transform (Hahn 1996).
a. Case 1: Uniform background
This is a reference case mainly for comparison with later examples. The background quantities in (7) are uniform, with U′ = 10 m s−1, N = 0.02 s−1, and H = 6 km. Figure 2 shows the vertical velocity w(x, z). Ray paths computed from (15) are superimposed on the plot with the thick line indicating the ray with k = −K0, the wavenumber of maximum |W(k)|.
Case 1: the vertical velocity w(x, z)/W0 in a uniform background, with the tsunami vertical displacement h(x) indicated at the bottom and ray paths superimposed. The ray path for k = −K0 is indicated by the thick line.
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
Geometrical spreading is very weak in this example, as illustrated by Fig. 3. The local magnitude of the vertical velocity |w| is plotted as a function of x at two heights: z = 0 (dashed line) and z = 150 km. The latter is obtained from the Fourier solution [(5); solid line] and the ray-tracing solution (circles), which are in close agreement. The peak |w| near x = 300 km decreases with altitude owing to geometrical spreading, but only slightly, by about 3% from z = 0 to z = 150 km.
Case 1: the local magnitude of the vertical velocity |w|/W0 at z = 0 (dashed line) and at z = 150 km from the Fourier (solid line) and ray (circles) solutions.
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
The ray-tracing solution in Fig. 3 is calculated from the varying horizontal distance δ between adjacent rays in Fig. 2. Along the ray |w| ~ δ−1/2.
b. Case 2: Idealized wind jet
The top panel of Fig. 4 shows 
Case 2: (top) 
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
The bottom panel of Fig. 4 shows the turning-point heights as a function of k. Also shown for comparison is the shape of the tsunami spectrum |W(k)| (bottom of the plot) and the shape of the wind jet (dashed line at the left). Over most of the spectrum there are two turning points, which are located near 100- and 120-km altitudes. For k/K0 < −2.2 the number of turning points increases to four. For k/K0 < −2.95 the waves are evanescent at z = 0 and are not included in this calculation.
Figure 5 shows w(x, z)/W0 in the top panel. There is substantial partial reflection from the wind jet, revealed by the almost vertical orientation of the wave phases below the wind jet, representing a standing wave pattern. The upgoing and downgoing components of the wavefield, calculated from (8)–(11), are plotted respectively in the middle and bottom panels. These are calculated only in regions where ϵ < 1 in (12), leaving gaps in the plotted solution near the wind jet.
Case 2: (top) w(x, z)/W0, with U′(z) indicated at the left, h(x) indicated at the bottom, and ray paths superimposed. The contributions from (middle) upgoing and (bottom) downgoing waves are also shown.
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
Case 2: reflection R(k) and transmission T(k) coefficients and the tsunami spectrum 
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
At k = −K0, the reflection and transmission coefficients of Fig. 6 are both about 0.71. These values compare well with the spatial solution. At the top of the domain, there are only upgoing waves, and the maximum spatial solution |w|/W0 is about 0.67. At z = 0, the maximum |w|/W0 for the downgoing component is about 0.66.
It is useful to compare these results with the predictions of wave tunneling theory, with two turning points surrounding a single evanescent region. In the simplest case m2 is a quadratic function of z in the turning-point region and is vertically symmetric about the turning-point height (Fröman and Fröman 2002, their section 3.49).
For a thin evanescent layer in the limit μ → 0, the reflection and transmission coefficients are equal, with value 
Figure 7 shows the solution w(x, z)/W0 for the reflecting lower boundary condition (14). It has the same features as in the top panel of Fig. 5, which was computed with the nonreflecting lower boundary condition [(13)]. There is again a standing wave pattern below the turning points and transmitted waves above the turning points. The main difference is the overall amplitude of the solution. The peak magnitude for |w|/W0 is about 1.6 in the case of Fig. 5 and about 1.1 in the case of Fig. 7. This implies some destructive interference caused by the gravity waves reflected from the ocean surface.
Case 2: w(x, z)/W0, as in Fig. 5 (top), but for the reflecting lower boundary condition in (14).
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
As another test of the parameters for case 2, we increased the maximum wind jet speed to 100 m s−1, setting U1 = 90 m s−1 in (17), with all other parameters the same as before and with the lower boundary condition [(13)]. This increases the reflection coefficient, but the transmission coefficient is still relatively high. For k > −1.5K0 the range of T is 0.49–0.58. At this wind speed, the wind jet half-width in (17) has to be increased to L = 30 km to reduce the transmission coefficients to less than 0.1 for all k.
c. Case 3: The Sumatra tsunami
We now examine gravity waves from the tsunami following the Sumatra earthquake on 26 December 2004. Most of the tsunami energy propagated westward from Sumatra into the Indian Ocean (Titov et al. 2005). As in Hickey et al. (2009) and Occhipinti et al. (2006), we obtain background atmospheric profiles for 0300 UTC on the day of the earthquake, and for 0°, 85°E—a location in the Indian Ocean roughly 10° west of the earthquake epicenter.
The vertical profiles of density ρ(z) and wind velocity U′ are obtained, respectively, from the empirical models Mass Spectrometer and Incoherent Scatter Radar (MSIS) and Horizontal Wind Model (HWM), as described in Picone et al. (2002) and Drob et al. (2008). There are several versions of these models, dating back to the early 1990s. We use the latest versions NRLMSISE-00 and HWM07 (Drob et al. 2008). The wind model HWM07 is a major upgrade from earlier versions and has a completely revised specification of the migrating tides in the mesosphere and lower thermosphere. Therefore our wind profile U′(z) is different from that used by Hickey et al. (2009), which was derived from an earlier version known as HWM93.
The vertical profiles for U′, N, and H are shown in Fig. 8. The output from MSIS leads to large (and likely spurious) oscillations for the calculated N(z) in the lower troposphere; see also the oscillations in buoyancy period in Fig. 2 of Hickey et al. (2009). For our calculation, these oscillations were replaced by a representative tropospheric value of N = 0.01 s−1 for altitudes up to z = 15 km. The sudden change in the slope of N(z) near the equatorial tropopause at 15-km altitude generates some partial reflection, but the transmission is large, as indicated by the results below.
Case 3: vertical profiles of (top to bottom) U′, N, and H for the case of the Sumatra tsunami.
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
Figure 9 shows 
Case 3: (top) 
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
Figure 10 shows w(x, z)/W0 in the top panel. The standing wave pattern below ~100 km indicates significant partial reflection. There is also significant partial transmission. Above 100-km altitude the waves are well represented by the solution for upgoing waves, with phase lines tilting westward with increasing altitude. The upgoing (middle panel) and downgoing (bottom panel) components are computed from (10) and (11).
As in Fig. 5, but for case 3.
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
Figure 11 shows the vertical eigenfunctions 
Case 3: the real part of 
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
The vertical eigenfunctions pass through the evanescent regions without much decrease in amplitude. For example, note the curve farthest to the left near z = 80-km altitude. Above the turning point near z = 77 km, the curve seems to start to decay with height as an evanescent wave (m2 < 0), but this is shortly interrupted by the presence of another turning point near z = 83 km, above which the wave resumes vertical propagation (m2 > 0). Above 100-km altitude, there is a sharp reduction in the amplitude of some of the eigenfunctions owing to the absence of downgoing waves above the highest turning point, as shown in the bottom panel of Fig. 10.
Figure 12 shows the individual terms in m2 of (7). All terms are normalized by dividing by 
Case 3: the terms in m2 of (7) normalized by dividing by 
Citation: Journal of the Atmospheric Sciences 71, 9; 10.1175/JAS-D-13-0309.1
The term N2/U2 in (7) is generally dominant in gravity wave applications. For tsunami-generated gravity waves this term is relatively small because U = U′ − c includes the large phase speed c of the tsunami. For realistic conditions the other terms of m2 can locally dominate N2/U2 and cause m2 to change sign repeatedly with altitude, explaining the abundance of turning points. While the exact altitudes of the turning points may be sensitive to some of the terms in m2, particularly −Uzz/U, the occurrence of multiple turning points is not sensitive to these details based on the experiments we have performed so far.
Hickey et al. (2009) modeled gravity waves generated by the Sumatra tsunami and found that the background winds can strongly limit the ability of the gravity waves to propagate into regions of the thermosphere above about 200-km altitude. This wind filtering effect was explained by the reflection of the gravity waves. In the case of their Fig. 6, it appears that most of the wind filtering occurs above about 180-km altitude, and that by 250-km altitude the wind filtering causes more than an order of magnitude reduction in the gravity wave amplitudes.
Our model does not cover these high altitudes, but as in Hickey et al. (2009) we did not find strong wind filtering below the thermosphere. We computed gravity wave solutions using background profiles from the latest and earlier versions of MSIS and HWM, for different times of the day during the Sumatra tsunami, and we also tested the effects of a strong wind jet (100 m s−1 peak speed) just above the mesopause region, as in Fig. 3 of Hickey et al. (2009). In all of these cases, the evanescent layers were sufficiently localized in the vertical to allow significant wave transmission. For altitudes below 150 km, partial reflection accounted for at most a factor of about 2 reduction in the upgoing wave amplitudes.
Above 200 km, the wind filtering as a result of turning points is likely to be stronger because the vertical scales of the background are much longer; see Fig. 3 of Hickey et al. (2009), particularly the east–west component of the wind, which is responsible for most of their wind filtering. The larger scales result in wider evanescent layers above turning points and a potentially greater exponential decay of the wave amplitudes with altitude.
6. Discussion
We considered two cases with background atmospheres that vary with height: an idealized wind jet centered near the mesopause and a realistic background for the tsunami following the Sumatra earthquake. In both cases all of the computed vertical eigenfunctions were evanescent for part of the way up to the thermosphere, with alternating layers of propagating (m2 > 0) and evanescent (m2 < 0) waves. In the Sumatra case, there were at least eight turning points for each vertical eigenfunction, with four intervening evanescent layers.
We found comparable magnitudes for the reflection and transmission coefficients, defined in (18) and plotted in Figs. 6 and 9. Two turning-point theory for a thin evanescent layer also gives similar magnitudes for the reflection and transmission coefficients, as noted in section 5. We modeled tsunami-generated gravity waves up to 180-km altitude. At higher altitudes, the vertical scales of the background variations are larger, as in the profiles of buoyancy period and wind in Figs. 2 and 3 of Hickey et al. (2009). This can result in wider evanescent layers and stronger partial reflection.
We computed spatial ray paths for the gravity waves. The ray corresponding to the peak amplitude of the tsunami spectrum (indicated by the thick curves in Figs. 2, 5, and 10) tracked the largest gravity wave amplitudes reasonably well. It may be of interest to continue the ray paths to higher altitudes with a form of the dispersion relation that includes dissipative terms for the thermosphere (Vadas and Fritts 2009; Vadas et al. 2009).
We ignored dissipation, the effects of a compressible atmosphere, and the reflection of downgoing waves from the ocean surface (except for the case of Fig. 7). We also omitted Fourier components that are evanescent at the ocean surface, even though they may become propagating a short distance above the ocean surface. The present results suggest that there would be significant wave transmission through a near-surface evanescent layer caused, for example, by a locally small value of N as might occur in a well-mixed boundary layer.
Acknowledgments
This research was partially supported by the Office of Naval Research through the NRL 6.1 work unit “The Boundary Paradox.” We thank Prof. M. Hickey for an early discussion.
REFERENCES
Artru, J., V. Ducic, H. Kanamori, P. Lognonné, and M. Murakami, 2005: Ionospheric detection of gravity waves induced by tsunamis. Geophys. J. Int., 160, 840–848, doi:10.1111/j.1365-246X.2005.02552.x.
Berry, M. V., 2005: Tsunami asymptotics. New J. Phys., 7, 129, doi:10.1088/1367-2630/7/1/129.
Broutman, D., S. D. Eckermann, and J. W. Rottman, 2009: Practical application of two-turning-point theory to mountain-wave transmission through a wind jet. J. Atmos. Sci., 66, 481–494, doi:10.1175/2008JAS2786.1.
Brown, G., and B. Sutherland, 2007: Internal wave tunnelling through non-uniformly stratified shear flow. Atmos.–Ocean, 45, 47–56, doi:10.3137/ao.v450104.
Drob, D. P., and Coauthors, 2008: An empirical model of the Earth’s horizontal wind fields: HWM07. J. Geophys. Res.,113, A12304, doi:10.1029/2008JA013668.
Einaudi, F., and C. O. Hines, 1970: WKB approximation in application to acoustic–gravity waves. Can. J. Phys., 48, 1458–1471, doi:10.1139/p70-185.
Fröman, N., and P. O. Fröman, 2002: Physical Problems Solved by the Phase-Integral Method. Cambridge University Press, 214 pp.
Hahn, S. F., 1996: Hilbert Transforms in Signal Processing. Artech House, 460 pp.
Heading, J., 1962: An Introduction to Phase-Integral Methods. John Wiley, 160 pp.
Hickey, M., G. Schubert, and R. Walterscheid, 2009: Propagation of tsunami-driven gravity waves into the thermosphere and ionosphere. J. Geophys. Res.,114, A08304, doi:10.1029/2009JA014105.
Inverarity, G. W., 2003: Numerically inverting a class of singular Fourier transforms: Theory and application to mountain waves. Proc. Roy. Soc. London, 459A, 1153–1170, doi:10.1098/rspa.2002.1076.
Lee, M. C., and Coauthors, 2008: Did tsunami-launched gravity waves trigger ionospheric turbulence over Arecibo? J. Geophys. Res.,113, A01302, doi:10.1029/2007JA012615.
Liu, J.-Y., Y.-B. Tsai, K.-F. Ma, Y.-I. Chen, H.-F. Tsai, C.-H. Lin, M. Kamogawa, and C.-P. Lee, 2006: Ionospheric GPS total electron content (TEC) disturbances triggered by the 26 December 2004 Indian Ocean tsunami. J. Geophys. Res.,111, A05303, doi:10.1029/2005JA011200.
Lovett, R., cited 2010: Tsunamis leave ionosphere all shook up. Nature News. [Available online at http://www.nature.com/news/2010/100914/full/news.2010.467.html.]
Nault, J. T., and B. R. Sutherland, 2007: Internal wave transmission in nonuniform flows. Phys. Fluids, 19, 016601, doi:10.1063/1.2424791.
Occhipinti, G., P. Lognonné, E. A. Kherani, and H. Hébert, 2006: Three-dimensional waveform modeling of ionospheric signature induced by the 2004 Sumatra tsunami. Geophys. Res. Lett., 33, L20104, doi:10.1029/2006GL026865.
Occhipinti, G., E. A. Kherani, and P. Lognonné, 2008: Geomagnetic dependence of ionospheric disturbances induced by tsunamigenic internal gravity waves. Geophys. J. Int., 173, 753–765, doi:10.1111/j.1365-246X.2008.03760.x.
Occhipinti, G., P. Coïsson, J. J. Makela, S. Allgeyer, E. A. Kherani, H. Hébert, and P. Lognonné, 2011: Three-dimensional numerical modeling of tsunami-related internal gravity waves in the Hawaiian atmosphere. Earth Planets Space, 63, 847–851, doi:10.5047/eps.2011.06.051.
Peltier, W. R., and C. O. Hines, 1976: On the possible detection of tsunamis by a monitoring of the ionosphere. J. Geophys. Res., 81, 1995–2000, doi:10.1029/JC081i012p01995.
Picone, J. M., A. E. Hedin, D. P. Drob, and A. C. Aikin, 2002: NRLMSISE-00 empirical model of the atmosphere: Statistical comparisons and scientific issues. J. Geophys. Res., 107, 1468, doi:10.1029/2002JA009430.
Simard, A., and W. Peltier, 1982: Ship waves in the lee of isolated topography. J. Atmos. Sci., 39, 587–609, doi:10.1175/1520-0469(1982)039<0587:SWITLO>2.0.CO;2.
Smith, R., 1979: The influence of mountains on the atmosphere. Advances in Geophysics, Vol. 21, Academic Press, 87–230, doi:10.1016/S0065-2687(08)60262-9.
Titov, V., A. B. Rabinovich, H. O. Mofjeld, R. E. Thomson, and F. I. González, 2005: The global reach of the 26 December 2004 Sumatra tsunami. Science, 309, 2045–2048, doi:10.1126/science.1114576.
Vadas, S. L., and D. C. Fritts, 2009: Reconstruction of gravity wave field from convective plumes via ray tracing. Ann. Geophys., 27, 147–177, doi:10.5194/angeo-27-147-2009.
Vadas, S. L., J. Yue, C.-Y. She, P. A. Stamus, and A. Z. Liu, 2009: A model study of the effects of winds on concentric rings of gravity waves from a convective plume near Fort Collins on 11 May 2004. J. Geophys. Res.,114, D06103, doi:10.1029/2008JD010753.
