astro-ph0608204/ms.tex
1: %{
2: % 02.03.06  modified by Dali
3: % 01.03.06  corrected by Sasha
4: % corrected by Aake
5: % corrected by Sasha
6: % corrected by Junwei
7: % 07.16.06  modified by Bob
8: % $Id: td.tex,v 1.22 2006/08/04 21:05:48 bob Exp $
9: 
10: \documentclass[12pt,preprint]{aastex}
11: \usepackage{amsmath}
12: %\def\apj{{ApJ}}
13: %\def\mnras{{MNRAS}}
14: %\def\solphys{{Solar Physics}}
15: %\def\aap{{A\&A}}
16: \bibliographystyle{aabib}
17: 
18: \newcommand{\oh}[1]{\omit\hidewidth #1\hidewidth}
19: 
20: 
21: \begin{document}
22: 
23: \title{Local helioseismology and correlation tracking
24: analysis of surface structures in realistic
25: simulations of solar convection}
26: 
27: 
28: \author{Dali Georgobiani, Junwei Zhao, Alexander G. Kosovichev}
29: \affil{Hansen Experimental Physics Laboratory, Stanford
30:        University, Stanford, CA, 94305}
31: %\email{junwei@quake.stanford.edu}
32: %\email{sasha@quake.stanford.edu}
33: 
34: \author{David Benson, Robert F. Stein}
35: \affil{Physics and Astronomy
36:        Department, Michigan State University, East Lansing, MI 48824}
37: %\email{bensond4@msu.edu}
38: %\email{bob@steinr.pa.msu.edu}
39: 
40: \author{{\AA}ke Nordlund}
41: \affil{Niels Bohr Institute, Copenhagen University,
42:        Juliane Maries Vej 30, DK--2100 K{\o}benhavn {\O}, Denmark}
43: %\email{aake@astro.ku.dk}
44: 
45: \keywords{convection---Sun: oscillations---methods: numerical}
46: 
47: \begin{abstract}
48: %\input{abstract}
49: We apply time-distance helioseismology, local correlation tracking
50: and Fourier spatial-temporal filtering methods to realistic
51: supergranule scale simulations of solar convection and compare the
52: results with high-resolution observations from the SOHO Michelson
53: Doppler Imager (MDI).  Our objective is to investigate the surface
54: and sub-surface convective structures and test helioseismic
55: measurements.  The size and grid of the computational domain are 
56: sufficient to resolve various convective scales from granulation 
57: to supergranulation. 
58: The spatial velocity spectrum is approximately a power law for scales
59: larger than granules, with
60: a continuous decrease in velocity amplitude with increasing size.
61: Aside from granulation no special scales exist, although a
62: small enhancement in power at supergranulation scales can be seen.
63: We calculate the time-distance diagram for $f$- and $p$-modes
64: and show that it is consistent with the SOHO/MDI observations.
65: From the simulation data we calculate travel time maps for surface 
66: gravity waves ($f$-mode).  We also apply correlation tracking to 
67: the simulated vertical velocity in the photosphere to calculate the
68: corresponding horizontal flows.  We compare both of these to the
69: actual large-scale (filtered) simulation velocities.
70: All three methods reveal similar large scale convective patterns and
71: provide an initial test of time-distance methods.
72: 
73: \end{abstract}
74: 
75: 
76: \section{Introduction}
77: %\input{intro}
78: \label{intro} Local helioseismology uses the observed oscillation
79: signals to probe the near surface structure of the Sun to determine
80: sound speed, flow velocities and magnetic field structures. One of
81: several methods used in this field is called the time-distance
82: helioseismology. It measures the time taken by a wave packet to
83: travel from one point on the solar surface to another
84: \citep{1993Natur.362..430D, 1997scor.proc..241K}. The travel time
85: for a wave depends on the wave speed and the flow
86: velocities along the ray paths. These effects can be separated by
87: measuring the wave travel time for waves propagating in opposite directions
88: along the same ray paths. Inversion methods applied to infer the
89: properties of subsurface convection and structures are based on
90: various approximations, such as a geometric acoustic ray approximation
91: \citep{2000SoPh..192..159K, 2001ApJ...557..384Z}, Fresnel-zone
92: approximation \citep{2001ApJ...553L.193J, 2004ApJ...607..554C}
93: and Born-approximation \citep{2002ApJ...571..966G}.
94: Unfortunately, until now, there has been no direct verification of
95: these methods \citep{2003soho...12..319J,2004soho...14..172W}.
96: 
97: Realistic 3D simulations of solar convection
98: \citep{{2000SoPh..192...91S},{Benson2006}} provide a way to
99: evaluate the accuracy and consistency of time-distance and other
100: local helioseismology techniques. These simulations provide a
101: model of convective motions in the upper convection zone
102: and possesses a rich spectrum of $f$- and $p$-mode oscillations excited
103: by the turbulent convection.  They can be used for
104: testing helioseismology methods. Therefore, it is very important
105: to apply some of the existing local helioseismology methods to the
106: simulated solar convection, and check whether the results are
107: consistent with the subsurface properties present in
108: the simulated flows. In addition, the comparison of the
109: oscillation properties obtained from the numerical model, such as
110: the  velocity spectrum, $k-\omega$ diagram and time-distance
111: (cross-covariance) diagram, with observations provides validation
112: for the simulation model.
113: 
114: Previously, the realistic simulations of solar convection have been
115: used to study the excitation mechanism of solar and stellar
116: oscillations \citep{2004SoPh..220..229S}, and some properties of
117: oscillation modes, such as the line asymmetry
118: \citep{2000ApJ...530L.139G}. However, these simulations were carried
119: out for small, granulation-size, computational domains. Recently,
120: with the progress of parallel supercomputing it became possible to
121: substantially expand the computational domain, both horizontally and
122: in depth, and to simulate the multi-scale solar convection, from
123: granulation to supergranular scales \citep{Benson2006}. These
124: simulations allow us for the first time to investigate the
125: properties of large-scale convection by applying helioseismology and
126: other methods used to analyze solar observations, and thus test
127: these methods and also evaluate how close the realistic simulations
128: are to the real Sun. Such tests are particularly important for the
129: local helioseismology methods, which are based on simplified models
130: of wave propagation.
131: 
132: The goal of this paper is to investigate the basic helioseismic
133: properties of the large-scale 3D simulations (the oscillation power
134: spectrum and wave travel times) and compare these with the
135: high-resolution observations from SOHO/MDI
136: \citep{1995SoPh..162..129S}.  An initial time-distance 
137: analysis is carried out for surface gravity waves ($f$-mode), which
138: are well suited for studying the surface and sub-surface structure
139: of solar convection on supergranulation scales
140: \citep{2000SoPh..192..177D}. Our analysis shows that travel times
141: from the simulation agree very well with travel times from the
142: SOHO/MDI observations. This demonstrates that the simulated data are
143: sufficiently close to the observed solar data, providing ground for
144: further detailed helioseismic measurements and inversions.
145: Large-scale structures can be detected in the simulated data set
146: using $f$-mode time-distance and local correlation tracking
147: techniques and are observed directly in filtered flow fields.
148: 
149: 
150: \section{Numerical Model and Observed Data} \label{model}
151: We study the results of a 3D, compressible, radiative-hydrodynamic
152: (RHD) code simulating the upper solar convection zone and
153: photosphere. The code calculates LTE, non-gray radiation transfer
154: and employs a realistic equation of state and opacities. More
155: description of the code is found in
156: \citet{{2000SoPh..192...91S},{Benson2006}}.
157: 
158: The computational domain of our simulations spans 48 Mm $\times$ 48
159: Mm horizontally and 20 Mm vertically, with a horizontal resolution
160: of 100 km and vertical resolution of 12 to 75 km.  The three
161: velocity components measured at 200 km above $\tau_{\rm cont} = 1$
162: are saved every minute. Their spatial grid is 500 by 500 pixels. The
163: domain is sufficiently large, and the time sequence of several hours
164: is long enough to obtain sufficient signal/noise by temporal
165: averaging to perform a helioseismic time-distance analysis and seek
166: evidence for the presence of large-scale flows. Eventually, the
167: depth dependence and dynamics of these large-scale flows will be
168: studied as well. We compare some of our findings with results
169: obtained from SOHO/MDI high resolution observations. This
170: observational data set is an 8.5 hour time series, with 1 minute
171: cadence, of a 211.5 Mm $\times$ 211.5 Mm horizontal patch of the MDI
172: Doppler velocity, on the spatial grid of 512 by 512 pixels with
173: the resolution of about 400 km per pixel. The solar rotation is
174: removed. Both sets of data are processed identically, whenever a
175: comparison between the observations and simulations occurs.
176: 
177: 
178: \section{Surface Structures}
179: %\input{struct}
180: \label{structures}
181: 
182: Oscillation modes in the simulations are excited naturally, due to
183: convective motions and realistic cooling at the solar surface
184: \citep{2000ApJ...530L.139G,2001ApJ...546..585S,2004SoPh..220..229S}.
185: We use the vertical velocity component from the simulation data as a
186: proxy for the observed Doppler velocity, and calculate the
187: power spectrum, $P(k,\omega)$, in the horizontal
188: wavenumber-frequency domain. Following the helioseismology
189: convention, we represent the power spectrum in terms of the
190: spherical harmonic degree, $\ell=kR_\sun$, where $R_\sun$ is the
191: solar radius, and the cyclic frequency, $\nu=\omega/2\pi$.
192: 
193: 
194: The power spectrum ($\ell-\nu$ diagram) calculated from the
195: simulated vertical velocity clearly shows a resolved $f$-mode along
196: with a $p$-mode spectrum that looks very similar to the results from
197: the SOHO/MDI Doppler images (Fig.~\ref{KW}). These simulations
198: possess a richer mode spectrum than in our smaller (6 Mm wide by 3
199: Mm deep) simulations because the wider and deeper domain encompasses
200: many more resonant modes within it.  The spectral resolution
201: achieved in these simulations allows us to test time-distance
202: helioseismology measurements
203: \citep{{1993Natur.362..430D},{1997SoPh..170...63D}}.
204: %%BOB: is it spectral resolution or signal to noise
205: 
206: \begin{figure}[ht]
207: %   \plotone{fig/both_power.l-nu.ps}
208:     \plotone{f1.ps}
209:     \caption{
210:     The power spectra ($\ell-\nu$ diagrams) for the simulated vertical velocity (left)
211:     and the Doppler velocity from the MDI high-resolution observations
212:     (right). The dark curve represents the theoretical $f$-mode.
213:     \label{KW}}
214: \end{figure}
215: 
216: Mode filtering is an essential part of the time-distance technique.
217: The convection signals (broad wedge in the lower right part of the
218: $\ell-\nu$ diagram) need to be removed. 
219: We construct the time-distance diagram for both
220: simulated and observed data by computing cross-covariance of
221: oscillation signals separated by a certain distance (Fig.~\ref{TD})
222: \citep{1997SoPh..170...63D}, and find a remarkable similarity
223: between them. Because of the higher resolution, in the
224: numerical simulation the cross-covariance signal extends to much
225: shorter distances than in the MDI data. This illustrates a
226: potential for small-scale helioseismic diagnostics of observations
227: of higher than MDI resolution, such as anticipated from the Solar-B
228: mission \citep{2004ASPC..325...87S}.
229: 
230: We show in section \ref{fmode} that the travel times from simulation
231: data agree with travel times from the SOHO/MDI observations very
232: well. Both the $\ell-\nu$ and time-distance diagrams illustrate the
233: potential of the simulations for testing helioseismology measurement
234: and inversions procedures.
235: 
236: \begin{figure}[!ht]
237: %       \plotone{fig/td_simu_real2.ps}
238:         \plotone{f2.ps}
239:         \caption{
240:     Time-distance diagrams for the simulated vertical velocity (left)
241:     and the Doppler velocity from the high-resolution MDI observations
242:     (right). The dark curve is time-distance relation computed from a
243:     standard solar model. The second bounce can be seen in both pictures.
244:         \label{TD}}
245: \end{figure}
246: 
247: In the following sections, we explore the surface structures in
248: the simulated data,  directly by averaging the flow field and using
249: spectral analysis (Sect.~\ref{direct}), and also by implementing the $f$-mode
250: time-distance analysis (Sect.~\ref{fmode}), and local correlation
251: tracking technique (Sect.~\ref{lct}).
252: 
253: 
254: \subsection{Fourier Analysis of Surface Velocities}
255: %\input{direct}
256: \label{direct}
257: 
258: We analyze the simulated velocity fields by calculating
259: spatial power spectra for both vertical and horizontal velocity
260: components and compare these with the spatial
261: power spectrum for the MDI Doppler signal.
262: %%%%%%%%%%%%%%%%
263: \begin{figure}[!t]
264: %       \plotone{fig/kw_noaver_20hr_BW.ps}
265:         \plotone{f3.ps}
266:         \caption{
267:     Spatial spectra for horizontal (dotted) and vertical (dashed) components of the
268:     simulated velocity, and the MDI Doppler signal (solid). Spectra are
269:     calculated for individual snapshots and then time-averaged, the
270:     simulation data over 20 hours and the MDI data over 8.5 hours.
271:         \label{SS}}
272: \end{figure}
273: %%%%%%%%%%%%%%%%
274: Power spectra are calculated for each time and are then averaged
275: over time.  This reduces the statistical fluctuations that 
276: are present in spectra from a single time.
277: In Fig.~\ref{SS} we show $V(\ell) = (\ell \, P(\ell))^{1/2}$, where 
278: $V(\ell)$ is velocity amplitude in the Fourier domain at spherical 
279: harmonic degree $\ell$ and $P(\ell)$ is time averaged power (velocity
280: squared) per unit $\ell$. There is a continuous decrease in the
281: velocity amplitude from granules ($\sim$ 1.5 Mm, $\ell \sim$ 2000-3000) 
282: to larger scales. Aside from
283: granulation no special scales are revealed in the spectra, although
284: there may be a small enhancement in the horizontal velocity ($V_{\rm
285: horiz}$) spectrum at supergranulation scales ($\ell \sim 200$).
286: 
287: In Fig.\ \ref{V_conv_osc} we separate motions into `convective' (dashed
288: lines)
289: and `oscillatory' (dotted lines) parts, according to whether the power in Fourier 
290: space lies above or below the line $\omega = c k$, where 
291: $c=6$ km\,s$^{-1}$ is approximately equal to the sound speed 
292: at the photosphere.  As can be seen from Fig.~\ref{KW} this provides
293: a rather clean split; the corresponding line in Fig.~\ref{KW}
294: would intersect the right hand side $\nu$-axis at $\nu\approx3.8$ mHz.
295: 
296: As shown by Fig.\ \ref{V_conv_osc}, the horizontal velocities (blue
297: curves) at the height of the MDI Ni $\lambda$ 676.78 nm line formation
298: (about 200 km above $\tau_{\rm 500} = 1$) are almost exclusively
299: of convective origin.  The vertical velocities (green curves), on the other hand,
300: are a mixture of convective (dashed line) and oscillation (dotted line) motions where the
301: convective motions are dominant at smaller scales and the oscillatory
302: motions are dominant at larger scales.
303: 
304: For granular and larger scales the
305: horizontal velocities are larger than the vertical velocities and
306: become increasingly dominant as the scale increases.  The reason that 
307: large scale horizontal motions are present at these heights is that
308: the atmosphere is not far from hydrostatic equilibrium and the pressure
309: fluctuations that exist below the surface to drive the large scale
310: horizontal flows imprint their fluctuations on the surface and produce
311: smaller amplitude large scale horizontal flows there also 
312: \citep{1982A&A...107....1N}.
313: 
314: Because of mass conservation, vertical convective velocities decrease 
315: more rapidly with size than the horizontal ones, with the ratio of
316: vertical to horizontal velocity amplitudes approximately inversely
317: proportional to size (cf.\ the green and blue dashed lines in 
318: Fig.\ \ref{V_conv_osc}).  At granular scales, where the vertical
319: velocity peaks, the vertical convective velocity at the height of 
320: formation of the $\lambda$ 676.78 nm line is still about 3--4 times
321: weaker than the horizontal one.
322: 
323: %%%%%%%%%%%%%%%%
324: \begin{figure}[!t]
325: %       \plotone{fig/V_conv_osc.ps}
326:         \plotone{f4.ps}
327:         \caption{Velocity spectra for the simulation (vertical = green, 
328: 	  horizontal = blue) and MDI observations (red) separated into 
329: 	  convective (dashed lines), oscillatory (dotted lines)
330:       and total (solid lines) components. The oscillatory signal is negligible in
331:       the horizontal velocity but dominates the vertical velocity and
332:       MDI Doppler velocity on large scales ($\ell \le 1000$).
333:       \label{V_conv_osc}}
334: \end{figure}
335: %%%%%%%%%%%%%%%%
336: 
337: In general, the MDI Doppler signal is a mix of horizontal and
338: vertical velocities. There is no obvious way to decouple these
339: components, although, depending on where on the solar disc the data
340: patch was located, one can estimate the approximate contribution of
341: the horizontal and vertical components. The current dataset was taken
342: near solar disk center (which at the time was about 7.5 heliographic 
343: degrees below the solar equator).  Nevertheless, because of the 
344: significant extension of the patch, the root-mean-square projection
345: of horizontal velocities into the line-of-sight was about 15\%,
346: while almost 99\% of the vertical velocity is projected onto
347: the line-of-sight. Indeed, if we combine 15\% of the
348: simulated horizontal velocities and 99\% of the simulated vertical
349: velocities, multiply by the instrument Modulation Transfer Function
350: (MTF) \citep{2001ApJ...548L.103W} and compare the resulting power
351: spectrum with the MDI power spectrum in Fig.~\ref{kw_MTF}, we get a
352: good correspondence, although the simulated power at large
353: scales (mostly oscillatory according to Fig.\ \ref{V_conv_osc}) falls 
354: somewhat short of the observed power.  Given that the numerical resolution
355: (96 km) is somewhat marginal on granular scales, where solar oscillations
356: are excited \citep{2001ApJ...546..585S}, this is not surprising.
357: 
358: From the slope of the MDI convective velocity (red dashed) curve in Fig.~\ref{V_conv_osc} one 
359: surmises that the convective LOS velocity observed with MDI over this 
360: patch actually derives most of its contribution at large scales from 
361: the horizontal velocity field.  An enhancement at supergranular scales 
362: (15-55 Mm, $\ell \sim$ 80-300) is visible.
363: 
364: %%%%%%%%%%%%%%%%
365:  \begin{figure}[!ht]
366: %        \plotone{fig/kw_MTF_BW.ps}
367:          \plotone{f5.ps}
368:          \caption{Simulated line-of-sight (LOS) component
369:          of the surface velocity spectrum (dotted), convolved with
370:       the instrumental MTF (dashed), compared with the MDI
371:          high-resolution spectrum (solid).
372:        \label{kw_MTF}}
373:  \end{figure}
374: %%%%%%%%%%%%%%%%
375: 
376: Averaging the velocity fields over several hours
377: eliminates the oscillations and suppresses the small scale features
378: but does not eliminate them.  To remove the small spatial scales, it
379: is necessary, in addition, to apply a low-pass k-space filter, 
380: where all values outside the circle $\sqrt{(k_x^2+k_y^2)} < 0.1$ Mm$^{-1}$ 
381: are set to zero.
382: We use this procedure to reveal the large-scale patterns in both
383: vertical and horizontal velocity components. Fig.~\ref{VzPmap} (left
384: panel) is an example of such a pattern in the $V_z$ component. An
385: alternative, cleaner, procedure would be to filter entirely in Fourier space,
386: eliminating the oscillations by removing power for $\omega \geq c k$ 
387: (as discussed in connection with Fig.\ \ref{V_conv_osc} above) and 
388: then applying a low-pass filter in $k$.  Here we follow the first
389: procedure which has been generally used in time-distance helioseismology.
390: 
391: \begin{figure}[!ht]
392: %       \plotone{fig/vz_f_pmap_nof_BW.ps}
393:         \plotone{f6.ps}
394:         \caption{
395:         Left: Simulated large scale vertical velocity. The velocity is
396:       averaged over 8.5 hours and smoothed with a low-pass filter.
397:       Light is downflow and dark upflow.
398:       Right: $p$-mode power in the frequency interval 2-5 mHz.  No
399:       low-pass filter has been applied.  The oscillatory power is
400:       concentrated in the intergranular lanes and is weakest in the
401:       upflow centers of the large-scale structures.
402:         \label{VzPmap}}
403: \end{figure}
404: 
405: We also construct $p$-mode power maps by Fourier-transforming the
406: vertical velocity fields in time at each spatial point, and summing
407: the result over the frequency interval 2-5 mHz (Fig.~\ref{VzPmap},
408: right panel). The $p$-mode power is rather intermittent on small
409: scales, and is in fact concentrated in intergranular lanes, as
410: shown by \citet{2001ApJ...546..585S}.  A large scale modulation,
411: correlated with the large scale vertical velocity field, is also 
412: discernible.  Presumably the large scale modulation is caused by
413: the slightly larger velocity amplitudes present in intergranular
414: lanes inside supergranulation scale downdrafts.
415: 
416: \subsection{F-Mode Time-Distance Analysis}
417: %\input{fmode}
418: \label{fmode}
419: 
420: An important question is whether local helioseismology methods can recover the
421: flow pattern from the oscillatory component of the
422: surface velocity field in these simulations.
423: We applied the time-distance technique to the $f$-mode oscillation,
424: because $f$-mode is the surface gravity mode and thus contains information
425: about horizontal flows and structures near the surface.
426: Therefore, the time-distance results can be directly compared with the
427: surface properties. The $f$-mode analysis is commonly used in
428: helioseismology.
429: We isolate the $f$-mode in $k_x, \, k_y, \, \nu$ space by retaining 
430: only values along a band centered on the theoretical $f$-mode ridge 
431: and setting all other values to zero, and then perform the time-distance
432: analysis, constructing outgoing and ingoing travel time maps, mean travel
433: times and travel time differences. We calculate the horizontal divergence
434: $dV_x/dx + dV_y/dy$ from the simulation data and compare it to the travel
435: time differences map after a low-pass filtering (Fig.~\ref{OI}).
436: As shown by \citet{2000SoPh..192..177D}, the travel-time difference is 
437: roughly proportional to the horizontal flow divergence.
438: \begin{figure}[!ht]
439: %       \plotone{fig/oi_div_BW.eps}
440:         \plotone{f7.ps}
441:         \caption{
442:     Comparison of the simulated horizontal velocity divergence (left;
443:     converging flows are light, diverging flows are dark) with
444:     $f$-mode outgoing and ingoing travel time difference map (right;
445:     incoming waves are faster in light areas, while outgoing waves are
446:     faster in dark areas)
447:     after a low-pass filter (as in Fig.~\ref{VzPmap}) has been applied.  Travel time
448:     extrema are 50 s for outgoing and 73 sec for incoming waves.
449: \label{OI}}
450: \end{figure}
451: We see a good agreement in the large-scale structures, with
452: correlation coefficient 0.88. Calculating
453: east-west and north-south travel time differences, we obtain proxies
454: for horizontal velocity components $V_x$ and $V_y$. We compare these
455: time difference maps with the actual velocities (Fig.~\ref{NS} for the
456: North-South componenet, the other looks similar). After
457: low-pass filtering, there is a good qualitative agreement between the two,
458: with correlation coefficients 0.7 for $x$-component and 0.73 for $y$-component.
459: \begin{figure}[!ht]
460: %        \plotone{fig/ns_vy_BW.eps}
461:         \plotone{f8.ps}
462:         \caption{
463:         Comparison between the simulated horizontal velocity ($V_y$,
464: 	  left, light is north-bound and dark is south-bound) and
465:         $f$-mode north-south travel time difference map (right, light
466: 	  represents faster north-bound and dark represents faster 
467: 	  south-bound waves). A low-pass
468:         filtering has been performed here as well. Travel time extrema
469: 	  are 39 s for north-bound and 72 s for south-bound waves.
470: \label{NS}}
471: \end{figure}
472: It is worth mentioning that the time-distance technique measures an
473: average over a certain depth, therefore a very high correlation is
474: not to be expected when comparing the time-distance results to a
475: single depth data, particularly above the surface. The agreement
476: might have been better if one looked at a depth representative of
477: the $f$-mode travel paths ($f$-mode kinetic energy is concentrated 
478: in the 2 Mm just below the surface).
479: 
480: 
481: \subsection{Local Correlation Tracking}
482: %\input{lct}
483: \label{lct}
484: 
485: Another possibility for deducing the horizontal velocities is to apply
486: the local correlation tracking (LCT) technique to the vertical velocity.
487: The LCT is a cross-correlation method
488: \citep[e.g.][]{1988ApJ...333..427N} applied to a time series
489: of solar granulation images. The cross correlation is spatially localized (within a
490: certain window, usually a Gaussian); its time average presumably measures
491: horizontal displacements of the flows.
492: We apply the LCT technique to the simulated vertical velocity field
493: and compare the resulting horizontal velocity proxies with the simulated
494: horizontal velocity components.  
495: Fig.~\ref{Vy_LCT} shows the $V_y$ component.
496: Velocities determined by the LCT method are tightly correlated with the
497: simulation velocity, 
498: with correlation coefficients of 0.99 for both $V_x$ and $V_y$ components.  
499: However, the LCT amplitudes are a factor 1.8 smaller.
500: Application of a low-pass filter to the simulation velocities brings this 
501: ratio down to 1.5.  These results are consistent with conclusions of 
502: \citet{2001A&A...377L..14R}.  According to them, the horizontal velocity
503: proxies deduced from granular motions underestimate actual velocities
504: by a variable factor, from 2.1 at small scales to 1.6 at large scales.
505: \begin{figure}[!ht]
506: %   \plotone{fig/vy_lct_BW.eps}
507:     \plotone{f9.ps}
508:     \caption{
509:     Simulated horizontal velocity component $V_y$ (left), averaged
510:         over 8.5 hours, and $V_y$, obtained by LCT analysis (right), after
511:         low-pass filtering.
512:         \label{Vy_LCT}}
513: \end{figure}
514: 
515: Thus, the realistic simulation data also provide a tool for testing
516: and improving the local correlation tracking techniques.
517: 
518: \section{Discussion}
519: %\input{conc}
520: \label{conc}
521: 
522: The new large-scale realistic 3D radiative-hydrodynamic simulations
523: of the upper layers of solar convection provide an excellent
524: opportunity to validate various techniques, widely used in solar
525: physics and helioseismology for directly obtaining otherwise
526: inaccessible properties (subsurface flows and structures, etc). On the
527: other hand, these analysis techniques also help to examine how
528: realistic the simulations are. We have performed an initial analysis
529: of the simulated data and compared our results with the outcome of
530: the SOHO/MDI observations. The similarity between the simulated and
531: observed $\ell-\nu$ and time-distance diagrams demonstrates that the
532: simulations can be efficiently used to perform and validate local
533: helioseismology techniques. 
534: This agreement also reveals a potential for high-resolution
535: time-distance measurements.
536: We carried out $f$-mode time-distance
537: calculations and local correlation tracking to obtain horizontal
538: velocities. We compare them to the simulated horizontal flows and
539: find a good qualitative agreement. We see large-scale structures in
540: both actual horizontal velocities and their proxies. The results of
541: this investigation provide the basis for further detailed
542: helioseismic analyses of the simulated data, including various
543: scheme of measuring p-mode travel times, approximations for the
544: sensitivity kernels and inversion procedures.
545: 
546: {\bf Acknowledgements.}  DG is grateful for the support of the SOHO/MDI
547: group at Stanford University.  DB and RFS were supported by NASA
548: grants NNG04GB92G and NAG 512450 and NSF grants AST-0205500 and
549: AST-0605738.  The work of {\AA}N was supported by a grant from the
550: Danish Natural Science Research Council.  Computing time was provided
551: by the Danish Center for Scientific Computing, and by grants from
552: the NASA Advanced Supercomputing Division.
553: 
554: \bibliography{ms}
555: 
556: \end{document}
557: %}
558: