astro-ph0702391/ms.tex
1: \documentclass[12pt,preprint]{aastex}
2: 
3: \begin{document}
4: 
5: \title{Contamination by Surface Effects of Time-distance Helioseismic Inversions for Sound Speed Beneath Sunspots}
6: 
7: \author{Couvidat, S. \& Rajaguru, S. P.}
8: 
9: \email{couvidat@stanford.edu}
10: 
11: \affil{W.W. Hansen Experimental Physics Laboratory, Stanford 
12: University, Stanford, CA 94305-4085}
13: 
14: \received{updated 2/8/2007}
15: 
16: \shorttitle{Contamination by Surface Effects of Time-distance Helioseismic Inversions}
17: \shortauthors{Couvidat}
18: 
19: \begin{abstract}
20: Using Doppler  velocity data from  the SOI/MDI instrument  onboard the
21: SoHO spacecraft, we do time-distance helioseismic inversions for 
22: sound-speed perturbations beneath 16 sunspots observed in high-resolution  
23: mode. We clearly detect ring-like regions of enhanced sound speed beneath 
24: most sunspot penumbrae, extending from near the surface to depths of about
25: $3.5$ Mm. Due to their location and dependence on frequency bands of p-modes
26: used, we believe these rings to be artifacts produced by a surface signal
27: probably associated with the sunspot magnetic field. 
28: \end{abstract}
29: 
30: \keywords{Sun: helioseismology --- (Sun:) sunspots}
31: 
32: \section{INTRODUCTION}
33: 
34: Time-distance helioseismology (Duvall et  \mbox{al.} 1993) has a track
35: record of  significant findings in solar physics:  from the subsurface
36: structure  and flow  fields of  sunspots (\mbox{e.g.} Duvall  et  \mbox{al.} 1996;
37: Kosovichev et \mbox{al.} 2000; Zhao et \mbox{al.} 2001) to the mapping
38: of large-scale  subsurface flows (Haber et \mbox{al.}  2002), and even
39: to  the in-progress  far side  imaging (Zhao,  private communication).
40: These results are  based on the assumption that the acoustic wave travel
41: times are only marginally affected by the presence  of  magnetic
42: fields. Recent work in sunspot seismology, based largely on helioseismic
43: holography techniques, however, point to a dominant influence of surface
44: magnetic field on two related phenomena: 
45: the so-called showerglass effect (Lindsey \& Braun 2005), and
46: the  effect  of  an  inclined  magnetic field  (\mbox{e.g.}  Cally  et
47: \mbox{al.} 2003;  Schunker et \mbox{al.}  2005; and Schunker  \& Cally
48: 2006). A clear understanding of the origin of these surface effects
49: is still lacking, and recent results 
50: point to contributions from several distinct
51: sources: while Schunker et \mbox{al.}  (2005), and associated modelling
52: work (Schunker  \& Cally 2006), emphasize the importance of wave propagation
53: in inclined magnetic field to the azimuthally varying local
54: control correlation phases (Braun \& Lindsey 2000) over penumbral regions,
55: Rajaguru et \mbox{al.}  (2007) show that such wave propagation effects
56: could manifest through altered height ranges of spectral line
57: formation that affects Doppler shift measurements within sunspots. 
58: While such surface effects have been confirmed present in time-distance 
59: measurements too (Zhao \&  Kosovichev 2006), these latter authors 
60: conclude that they do not change appreciably the previously obtained
61: deeper extending two-region (cold-hot) structure of sound-speed perturbations
62: below a sunspot and thus stress the presence of real subphotospheric structures. 
63: Here we analyze  16 active  regions of  different sizes  and magnetic
64: field fluxes: we compute travel-time perturbation maps for 11 travel distances 
65: and invert them for 3-dimensional sound-speed perturbations beneath each active region. 
66: We show the presence of ring-like structures of significant positive sound speed changes,
67: down to a depth of about  $3.5$ Mm, that seem 
68: likely to be artifacts due to our ignoring the surface magnetic effects in the
69: inversion procedures. We perform experiments to check the origin of these surface
70: effects: the amplitude effect [interaction of phase-speed filter with
71: localised reduced amplitudes of oscillations within sunspots, Rajaguru et \mbox{al.} (2006)]
72: is shown not to account for it, the location and frequency dependence of the ring-like
73: sound speed excesses are found to be related to similar dependences of travel times observed recently
74: by Braun \& Birch (2006), and the ring-like structures are shown to be present only on the outgoing travel-time maps.
75: 
76: In section 2 we describe the methodology applied here to obtain travel-time
77: maps and to invert for the sound-speed perturbation. In section 3 we present
78: our results. We conclude in section 4.
79: 
80: \section{TIME-DISTANCE ANALYSIS}
81: \subsection{Datasets}
82: 
83: We use high-resolution dopplergrams derived from filtergrams of circular
84: polarization component of Ni {\sc i} ($\lambda$=6768 \AA) observed by the
85: Solar Oscillations Investigation/Michelson  Doppler Imager (SOI/MDI) onboard
86: SoHO (Solar and Heliospheric Observatory) (Scherrer et  \mbox{al.} 1995).
87: These  dopplergrams are mapped  using a Postel  projection and
88: are  tracked  at  the  Snodgrass  differential rotation  rate corresponding
89: to central latitude of the MDI hi-res field of view (mapping  and  tracking
90: are done using the ``fastrack'' program in the Stanford MDI/SOI pipeline software).
91: Dopplergrams  that  are  contiguous in  time  are  stacked
92: together   to  produce  3D   datacubes  $\phi(\mathbf{r},t)$   of  the
93: line-of-sight (l.o.s) Doppler   velocity  at   the   solar  surface,   where
94: $\mathbf{r}=(x,y)$  is the  horizontal  vector, and  $t$  is the  time
95: coordinate.     The    dimensions     of    these     datacubes    are
96: $256\times256\times512$ nodes, with a spatial resolution $dx=dy=0.826$
97: Mm, and  a temporal  resolution $dt=1$ minute.  The same  procedure is
98: applied  to produce  datacubes of  the l.o.s magnetic
99: field  $\mathbf{B}(\mathbf{r},t)$,  and  of  the  continuum  intensity
100: $I_0(\mathbf{r},t)$. The  continuum intensity is the  intensity in the
101: solar continuum as measured by MDI at the tuning position number zero,
102: corresponding to two symmetrical measurements 150 m\AA $\,$ away from the
103: Ni {\sc i} line center.  The Doppler velocity  datacubes are detrended
104: by a  linear fit of the  long-term temporal trend.  When a dopplergram
105: has  no data or  is corrupted,  we replace  it by  the average  of its
106: neighbors.
107: 
108: We have selected datacubes containing sunspots  of various  sizes  and magnetic  field
109: strengths. Unfortunately, due to a saturation effect in the MDI magnetogram
110: measurements in large sunspots, the absolute value of magnetic field strength
111: in the umbrae of such spots is smaller than the actual values present in them.
112: This phenomenon was explained by Liu, Norton, \& Scherrer (2007) as being due to  a
113: limitation in the onboard processing of dopplergrams and magnetograms.
114: It limits our access to very large sunspots.
115: 
116: For the purpose of  this paper, a  sunspot is  defined as a  group of
117: contiguous pixels with a l.o.s magnetic field strength larger than 150
118: Gauss and that exhibits both an umbra and penumbra. The area of a spot
119: is only the  number of pixels within this spot  multiplied by the area
120: of a pixel $dx \times  dy=0.682$ Mm$^2$. The l.o.s magnetic field flux
121: is the value of  $\mathbf{B}(\mathbf{r},0)$ at each pixel muliplied by
122: the area  of a pixel. Table \ref{table.spot} summarizes the properties
123: of the  active regions studied in this paper, while  Figure \ref{f1}
124: shows maps of their continuum intensity $I_0$.
125: 
126: \subsection{Travel-Time Estimation}
127: 
128: The time-distance formalism  was introduced  by Duvall  et \mbox{al.}
129: (1993). It is based on computation of  cross-covariances between
130: the solar oscillation signals at two locations on the solar surface (a
131: source  at $\mathbf{r_1}$  and a receiver at $\mathbf{r_2}$).  Due to  the
132: stochastic excitation of acoustic waves by the solar convection zone and
133: due to the oscillation signal at any location being a superposition of 
134: a large number of waves of different travel distances (i.e., of different
135: horizontal phase velocities $v_{ph}=\omega/k$, where $k$ is the horizontal
136: wavenumber and $\omega$ is the temporal angular frequency), the cross-covariances 
137: are very noisy and need to be phase-speed filtered and averaged
138: (Duvall  et \mbox{al.} 1997). The Doppler velocity data cube $\phi(\mathbf{r},t)$ is phase-speed
139: filtered in  the  Fourier  domain, using  a Gaussian filter $F(k,\omega;\Delta)$  
140: for each travel distance $\Delta$. The standard scheme of averaging is to 
141: average the point-to-point cross-covariances over an annulus of radius 
142: $\Delta=|\mathbf{r_2}-\mathbf{r_1}|$ centered on the source. Such point-to-annulus
143: cross-covariances are  computed for several distances  $\Delta$ (55 in
144: this paper),  and then averaged by  groups of 5 distances to further
145: increase  their signal-to-noise  ratio. A detailed explanation of all these 
146: steps in the analysis process can be found in, \mbox{e.g.}, 
147: Couvidat, Birch, \& Kosovichev (2006), along with a table
148: of distances $\Delta$  and phase-speed filter characteristics used
149: (see Table 1 of previous reference).
150: 
151: The  point-to-annulus  cross-covariances   are  fitted  by  two  Gabor
152: wavelets (Kosovichev \& Duvall 1997): one for the positive times, one
153: for the negative times. We use the ingoing (i) and outgoing (o) phase travel times  
154: $\tau_{\mathrm{i/o}}(\mathbf{r},\Delta)$ returned by the fits. The average of
155: these  two travel  times, $\tau_{\mathrm{mean}}(\mathbf{r},\Delta)$, is,
156: in   first   approximation,   sensitive   only   to   the   sound-speed
157: $c(\mathbf{r},z)$ in the  region of the Sun traversed  by the wavepacket
158: (where  $z$ is  the  vertical coordinate). Using  a solar  model as  a
159: reference  we can  relate the  mean travel-time  perturbations $\delta
160: \tau_{\mathrm{mean}}(\mathbf{r},\Delta)$  to the sound-speed
161: perturbations $\delta c(\mathbf{r},z)$ through an integral relation:
162: 
163: \begin{equation}
164: \mathrm{\delta} \mathrm{\tau_{\mathrm{mean}}}(\mathbf{r},\mathrm{\Delta}) = \int\!\!\!\!\int_S d\mathbf{r'} \int_{-d}^0 dz \, K(\mathbf{r}-\mathbf{r'},z;\mathrm{\Delta}) \, \frac{\mathrm{\delta} c^2}{c^2}(\mathbf{r'},z) \label{eq1}
165: \end{equation}
166: 
167: where  $S$ is  the  area of  the region,  and  $d$ is  its depth.  The
168: sensitivity kernel for  the relative squared sound-speed perturbations
169: is  given  by $K$.  Here  we  use  Born-approximation kernels kindly 
170: provided  by \mbox{A. C.}  Birch (Birch, Kosovichev, \&  Duvall 2004). These
171: kernels  take  into  account  the  finite wavelength  effects  of  the
172: wavefield. Equation 1 is only  approximate because effects on the mean
173: travel times other  than the  sound-speed perturbation  are completely
174: ignored. For instance, Bruggen \& Spruit (2000) describe the impact of
175: changes in the upper boundary  condition in sunspots due to the Wilson
176: depression; Woodard (1997) and  Gizon \& Birch (2002) demonstrate that
177: increased  wave damping  in  sunpots can  introduce  shifts in  travel
178: times;  finally,  the significant  magnetic  field effects,  already  mentioned  in
179: Introduction, also  contribute to $\delta \tau_{\mathrm{mean}}(\mathbf{r},\Delta)$.
180: 
181: \subsection{Inversion of Travel-Time Maps}
182: 
183: Based on Equation \ref{eq1} the mean travel-time  perturbation maps (11 of them  per datacube) are
184: inverted  for  sound-speed  perturbation   using  a  regularized
185: least-squares scheme  based on a modified Multi-Channel Deconvolution
186: algorithm (MCD; see Couvidat et \mbox{al.} 2005 for details on the modified algorithm, and
187: Jacobsen et \mbox{al.} 1999 for the basic algorithm). The
188: modified   algorithm    includes   both   horizontal    and   vertical
189: regularizations. We use the weighted norm of, and that of the first 
190: derivatives of, the solution $\mathrm{\delta} c^2/c^2(\mathbf{r},z)$ for
191: vertical and horizontal regularization, respectively.  Following 
192: Couvidat et \mbox{al.} (2005) the covariance matrix of
193: the  noise  in the  travel-time  maps  is  included in  the  inversion
194: procedure, and  is obtained by  40 realizations of  noise travel-time
195: maps based  on the noise  model by Gizon  \& Birch (2004).   Both the
196: horizontal and vertical regularization parameters are constant for the
197: inversion  of  travel-time   maps  of  all  our  datasets.  These
198: parameters are chosen  empirically. We invert for 13  layers in depth,
199: with the center of the deepest layer located $28.5$ Mm below the solar
200: surface.
201: 
202: \section{RESULTS AND DISCUSSION}
203: 
204: Figure \ref{f2}  shows the mean travel-time perturbation  maps  for
205: a travel distance $\Delta=11.6$ Mm over the 16 active regions. The main feature
206: visible  on most  maps is  a central  white disk  of  increased travel
207: times. Surrounding this disk, a dark ring corresponding to a decreased
208: mean travel time is clearly  present. Umbra and penumbra boundaries overplotted
209: on Figure \ref{f2} show that this dark ring is mostly located within the penumbral region of  the spot.
210: Such  dark  rings actually  appear in travel-time maps for 2  of the  11
211: distances  $\Delta$ studied  here: $\Delta=8.7$  Mm  and $\Delta=11.6$
212: Mm.        An        azimuthal        average        of        $\delta
213: \tau_{\mathrm{mean}}(\mathbf{r},\Delta)$ around the sunspot center for
214: 5 circular sunspots shows the average amplitude of the rings and their
215: extent  (see Figure  \ref{f9}).  The rings do  not  seem to  be present  at
216: $\Delta=6.2$  Mm; some  spots exhibit  them but  most do not.   
217: 
218: When travel-time maps  are inverted for sound-speed perturbations,
219: they  produce the  well known  two-region structure  below  the active
220: region: first the sound speed decreases just below the photosphere and
221: then increases  at a depth of  about 3-4 Mm, depending  on the sunspot
222: size  and magnetic  flux.  These two  regions  can be  seen in  Figure
223: \ref{f6}:  this is  a vertical  cut in  the inversion  result  for the
224: active region NOAA 8397.   The dark ring surrounding the corresponding
225: sunspot on the  travel-time maps of Figure \ref{f2}  appears on Figure
226: \ref{f6} as  two bright  arms extending from  the region  of increased
227: sound-speed toward the surface. These arms are especially well visible
228: at a  depth of  about $2$  Mm. Their amplitude  decreases when  we get
229: closer to the solar surface. The rings/arms are not visible anymore at
230: the  surface, in agreement  with the  travel-time map  at $\Delta=6.2$
231: Mm.   Such   ring-like/arm-like  structures   are   present  on   most
232: sunspots.  Figure  \ref{f5} shows  horizontal  cuts  in the  inversion
233: results for the layer $z=-2.38$ to $z=-1.42$ Mm, while Figure \ref{f8}
234: shows similar cuts  for the layer $z=-1.42$ to  $z=-0.62$ Mm.
235: 
236: An azimuthal  average of the  sound-speed perturbation for  5 circular
237: sunspots as  a function of the  radial distance to the  spot center is
238: plotted on Figure  \ref{f4}, for the layers $z=-2.38$  to $z=-1.42$ Mm
239: (upper  panel) and  $z=-1.42$ to  $z=-0.62$ Mm  (lower panel).  At the
240: layer  $z=-2.38$  to  $z=-1.42$  Mm,  the averaged  amplitude  of  the
241: increased sound-speed  region is always  above the noise level  on the
242: inverted results, with a  maximum value of about $\delta c^2/c^2=6\%$,
243: corresponding   to  a   sound-speed  increase   of  about   $0.45$  km
244: s$^{-1}$. Neglecting  the magnetic field influence, any  change in the
245: first adiabatic exponent, and any change in the mean molecular weight,
246: we have  $\delta c^2/c^2=\delta T/T$. Therefore, a $6\%$  increase in
247: $\delta c^2/c^2$,  translates into an upper limit  for the temperature
248: increase at $z=-1.9$ Mm of about  $1100$ K. For the layer $z=-1.42$ to
249: $z=-0.62$ Mm,  the maximum value  of the sound-speed  perturbation for
250: the 5 sunspots of Figure  \ref{f4} is about $\delta c^2/c^2=2\%$ after
251: azimuthal  averaging,   corresponding  to  an  upper   limit  for  the
252: temperature increase of $295$ K.
253: 
254: It is  worth mentioning that these  ring-like/arm-like structures have
255: already been detected  in the past: for instance,  Figure 4 of Hughes,
256: Rajaguru,  \& Thompson  (2005) clearly  shows a  ring on  the inverted
257: sound-speed  perturbation  map  at  $z=-2.3$  to  $z=-1.7$  Mm.  Their
258: inversion was performed using  ray-path approximation kernels, and the
259: algorithm  was LSQR  and not  the MCD  used here.  Another  example is
260: Figure 8  of Couvidat  et \mbox{al.} (2004)  which shows a  clear ring
261: structure  around   the  sunspot  NOAA  8243,   after  inversion  with
262: Fresnel-zone sensitivity kernels. The significance of these rings had been 
263: dismissed at that time.
264: 
265: To understand the nature of these rings ---artifacts or real physical features---
266: we perform several analyses. First, a recent study by Rajaguru  et \mbox{al.} (2006) shows that  the interaction of
267: phase-speed  filtering in  time-distance  analysis and  the spatially 
268: localized p-mode amplitude reduction in sunspots
269: introduces travel-time artifacts that are significant for short travel distances. 
270: To  make sure that the rings we observe
271: on the travel-time maps are not due to this p-mode amplitude effect, we applied
272: the power correction suggested  in their paper to a  couple of sunspots: the
273: dark ring is still present after correction, even though with a different size and a
274: different amplitude. Figure \ref{f11} shows the impact of this correction:
275: there might be a dependence on the sunpsot size; the dark ring being more 
276: affected by the power correction for small spots. 
277: Second, Braun  \& Birch (2006) observed frequency variations of solar p-mode 
278: travel times, using acoustic holography. They believe these variations are,
279:  at least partly, produced by surface effects due to magnetic fields.
280: To test whether or not the  dark rings have this frequency dependence, we
281: followed Braun  \& Birch (2006) and computed some  travel-time maps at  $\Delta=11.6$ Mm 
282: with the same phase-speed filter but different temporal frequency filters: 
283: on top of the phase-speed filter $F(k,\omega;\Delta)$ we add a Gaussian $\nu=\omega/(2\pi)$ 
284: filter with a FWHM of $\delta \nu=1$ mHz and centered on different frequencies ($\nu=$3, 4, and
285: $4.5$ mHz). The resulting travel-time maps for NOAA 9493 are shown in Figure \ref{f10}: 
286: the size and amplitude of the ring clearly depend on  the 
287: $\nu$ filter used. At low frequencies ($\nu$ filter centered on 3 mHz) the ring
288: is almost non-existent, but as the frequency content of the wave-packets shifts to higher
289: frequencies the ring broadens and its amplitude increases. 
290: At $4.5$ mHz, the ring extends inside the umbra and, maybe, even 
291: beyond the penumbra. Therefore, the near absence of rings on travel-time maps for 
292: $\Delta=6.2$ Mm might be partly due to the phase-speed filter we use at this distance: 
293: this filter selects wavepackets with a maximum power at about $\nu=3.5$ mHz. 
294: However, for $\Delta=11.6$ Mm, the selected wavepackets have a maximum power at about $4.25$ mHz.
295: It should be noticed that with our frequency and broad phase-speed filters,
296: selecting a specific frequency band also means selecting a slightly different average phase speed: both
297: effects are entangled. Therefore what we consider to be a dependence on frequency might partly
298: be a dependence on phase speed. Figure \ref{f10} also shows that the travel times in the umbra of the sunspot depend on frequency, 
299: albeit to a lesser extent than in the penumbra. Therefore, the rings might be the manifestation 
300: of a phenomenon affecting the entire spot, and directed from the periphery (the penumbra) to 
301: the center (the umbra) when the wave-packet frequency selected for the time-distance 
302: analysis is increased. Nevertheless, we do believe that 
303: the qualitative sound-speed inversion result below the umbra (two-region structure) holds, because 
304: the mean travel-time perturbation in the vicinity of the spot center is still consistently positive 
305: with the frequency filters applied for $\Delta=11.6$ Mm. However, this analysis undoubtedly confirms that the layers below 
306: a spot umbra have a lateral extent and an inverted $\delta c^2/c^2$ amplitude that depend on the 
307: wave-packet frequency. The problem seems less acute for deeper layers: the travel-time maps for 
308: larger distances $\Delta$ also vary with the frequency but no structure similar to the rings
309: is ever present. Even though we suggest that the qualitative result of the sound-speed inversion 
310: below a sunspot is valid, any attempt to quantitatively 
311: assess the corresponding sound-speed perturbation amplitude and the lateral extent of the decreased sound-speed region 
312: are inextricably connected to the frequency band selected. 
313: Third, with the phase-speed filters (no frequency filter), we notice a strong asymmetry between ingoing and outgoing 
314: travel-time maps: the rings seem to be present only on the outgoing travel times (see Figure \ref{f13}). 
315: This asymmetry might be an additional sign of an interaction of the wavefield with surface magnetic 
316: field (as suggested in, \mbox{e.g}, Lindsey \& Braun 2005), but such a feature can also be
317:  explained by a flow (as first suggested by Duvall et \mbox{al.} 1996). Figure \ref{f13} shows that the ring appears
318: both as partly due to a sound-speed perturbation (because it is visible on the mean travel-time maps),
319: and as partly due to a flow if we accept the interpretation of Duvall et \mbox{al.} (1996) (because it is visible 
320: on the difference travel-time maps). The sign of $\delta \tau_{\mathrm{o}}-\delta \tau_{\mathrm{i}}$
321: at $\Delta=11.6$ Mm points to an inflow below the spot umbra, and an outflow in the penumbral ring. Such a reversal in the 
322: flow direction seems physically implausible. Finally, Figure \ref{f12}  shows   a  loose  dependence  of   the  maximum  sound-speed
323: perturbation in these rings at a depth of $z=-2.38$ to $z=-1.42$ Mm as
324: a function of  the total magnetic flux of the  sunspot: the larger the
325: magnetic  flux or  the sunspot  size (Table  \ref{table.spot}  shows a
326: linear relation  between these two parameters), the  larger is $\delta
327: c^2/c^2$ in the arms/rings. This last observation is inconclusive though because
328: the subsurface structure of the spot might also depend on the total flux: it does not help
329: in discriminating between the ring-as-artifact and the ring-as-real-structure hypotheses.
330: 
331: Considering that the rings are located mainly below  the sunspot penumbra in the  inversion results and
332: inside  the penumbra  on the  travel-time maps, that there is a strong asymmetry 
333: between ingoing and outgoing travel-time maps (with rings present only on the outgoing travel times), 
334: and that the ring size and amplitude have a very clear frequency dependence, it seems hard to
335: believe that the sound-speed increase they imply is the signature of a
336: real solar  feature located in depth. Indeed  it seems more likely that
337: these  rings are spurious features  produced by frequency dependent surface effects
338: somewhat related to the strong magnetic field of the sunspots, which are not
339: accounted for in the inversion procedure.  
340: 
341: \section{CONCLUSION}
342: 
343: We clearly detect rings  of negative mean travel-time perturbations on
344: the  travel-time  maps of  most  sunspots  for  travel distances  between
345: $\Delta=8.7$  and $\Delta=11.6$  Mm. These  rings  produce significant
346: arm-like/ring-like  structures  on  the  inversion  results,  mimicking
347: regions  of increased sound speed.   However due  to the  locations of
348: these structures (immediately below  the sunspot penumbrae), and their
349: sensitivity  to   the  filtering  applied  to   the  Doppler  velocity
350: datacubes, it is likely  that they are artifacts produced by
351: the interaction  of the wavefield  and surface magnetic  fields. These
352: significant  structures were  not mentioned  in the  paper by  Zhao \&
353: Kosovichev (2006). Even though we agree with their conclusion that the
354: effect  of the  surface magnetic  field probably  does not  change the
355: qualitative inversion result of the basic sunspot structure (the two-region
356: structure), we show that these inversion results for the first 3-4
357: Mm below the solar surface  appear to be significantly contaminated by surface
358: effects probably of magnetic origin. Indeed, the rings could be the most visible
359: manifestation of a general phenomenon that also affects the amplitude and the lateral extent 
360: of the sound-speed perturbation in the shallower subsurface layers.
361: 
362: 
363: \section*{Acknowledgments}
364: 
365: This work  was supported by  NASA grants NNG05GH14G (MDI) \& NNG05GM85G (LWS). We
366: are grateful to A. C. Birch for providing us with his Born kernels, to
367: R. Wachter for fruitful discussions and  for teaching S. C. how to use
368: the fastrack software, to T. L. Duvall for his very useful advice, and to the anonymous
369: referee for improving this paper.
370: 
371: \begin{thebibliography}{}
372: \bibitem[Birch et al.(2004)]{Birch04} Birch, A. C., Kosovichev, A. G., \& Duvall, T. L., Jr. 2004, \apj, 608, 580
373: \bibitem[Braun \& Lindsey(2000)]{Braun00} Braun, D. C., \& Lindsey, C. 2000, \solphys, 192, 307
374: \bibitem[Braun \& Birch(2006)]{Braun06} Braun, D. C., \& Birch, A. C. 2006, \apj, 647, L187
375: \bibitem[Bruggen \& Spruit(2000)]{Bruggen00} Bruggen, M., \& Spruit, H. C. 2000, \solphys, 196, 29 
376: \bibitem[Cally et al.(2003)]{Cally03} Cally, P. S., Crouch, A. D., \& Braun, D. C. 2003, MNRAS, 346, 381
377: \bibitem[Couvidat et al.(2004)]{Couvidat04} Couvidat, S., Birch, A. C., Kosovichev, A. G., \& Zhao, J. 2004, \apj, 607, 554
378: \bibitem[Couvidat et al.(2005)]{Couvidat05} Couvidat, S., Gizon, L., Birch, A. C., Larsen, R. M., \& Kosovichev, A. G. 2005, ApJS, 158, 217
379: \bibitem[Couvidat et al.(2006)]{Couvidat06} Couvidat, S., Birch, A. C., \& Kosovichev, A. G. 2006, \apj, 640, 516
380: \bibitem[Duvall et al.(1993)]{Duvall93} Duvall, T. L., Jr., Jefferies, S. M., Harvey, J. W., \& Pomerantz, M. A. 1993, Nature, 362, 430
381: \bibitem[Duvall et al.(1996)]{Duvall96} Duvall, T. L., Jr., D'Silva, S., Jefferies, S. M., Harvey, J. W., \& Schou, J. 1996, Nature, 379, 235
382: \bibitem[Duvall et al.(1997)]{duvalletal97} Duvall, T. L., Jr. et al. 1997, Sol. Phys., 170, 63
383: \bibitem[Gizon \& Birch(2002)]{Gizon02} Gizon, L., \& Birch, A. C. 2002, \apj, 571, 966
384: \bibitem[Gizon \& Birch(2004)]{Gizon04} Gizon, L., \& Birch, A. C. 2004, \apj, 614, 472
385: \bibitem[Haber et al.(2002)]{Haber02} Haber, D. A., Hindman, B. W., Toomre, J., Bogart, R. S., Larsen, R. M., \& Hill, F. 2002, \apj, 570, 855 
386: \bibitem[Hughes et al.(2005)]{Hughes05} Hughes, S. J., Rajaguru, S. P., \& Thompson, M. J. 2005, \apj, 627, 1040
387: \bibitem[Jacobsen et al.(1999)]{Jacobsen99} Jacobsen, B. H., M$\o$ller, I., Jensen, J. M., \& Effers$\o$, F. 1999, Phys. Chem. Earth, 24, 15
388: \bibitem[Kosovichev et al.(1997)]{Kosovichev97} Kosovichev, A. G., \& Duvall, T. L., Jr. 1997, Proceedings of SCORe'96 : Solar Convection and Oscillations and their Relationship, Eds.: F.P. Pijpers, J. Christensen-Dalsgaard, and C.S. Rosenthal, Kluwer Academic Publishers (Astrophysics and Space Science Library Vol. 225), 241
389: \bibitem[Kosovichev et al.(2000)]{Kosovichev00} Kosovichev, A. G., Duvall, T. L., Jr., \& Scherrer, P. H. 2000, \solphys, 192, 159
390: \bibitem[Lindsey \& Braun(2005)]{Lindsey05} Lindsey, C., \& Braun, D. C. 2005, \apj, 620, 1107
391: \bibitem[Liu et al.(2007)]{Liu07} Liu, Y., Norton, A. A., \& Scherrer, P. H. 2007, \solphys, accepted
392: \bibitem[Rajaguru et al.(2006)]{Rajaguru06} Rajaguru, S. P., Birch, A. C., Duvall, T. L., Jr., Thompson, M. J., \& Zhao, J. 2006, \apj, 646, 543
393: \bibitem[Rajaguru et al.(2007)]{Rajaguru07} Rajaguru, S. P., Sankarasubramanian, K., Wachter, R., \& Scherrer, P. H. 2007, \apj, 654, L175
394: \bibitem[Scherrer et al.(1995)]{Scherrer95} Scherrer, P. H., et al. 1995, \solphys, 162, 129
395: \bibitem[Schunker et al.(2005)]{Schunker05} Schunker, H, Braun, D. C., Cally, P. S., \& Lindsey, C. 2005, \apj, 621, L149
396: \bibitem[Schunker \& Cally(2006)]{Schunker06} Schunker, H., \& Cally, P. S. 2006, MNRAS, 372, 551
397: \bibitem[Woodard(1997)]{Woodard97} Woodard, M. F. 1997, \apj, 485, 890
398: \bibitem[Zhao et al.(2001)]{Zhao01} Zhao, J., Kosovichev, A. G., \& Duvall, T. L., Jr. 2001, \apj, 557, 384
399: \bibitem[Zhao et al.(2006)]{Zhao06} Zhao, J., \& Kosovichev, A. G. 2006, \apj, 643, 1317
400: \end{thebibliography}
401: 
402: \clearpage
403: 
404: \begin{deluxetable}{ccccccccc}
405: \tablecaption{Characteristics of the Sunspots Studied, for a Threshold of 150 G \tablenotemark{a}\label{table.spot}}
406: \tablewidth{0pt}
407: \tablehead{\colhead{NOAA} & \colhead{Date} & \colhead{Area} & \colhead{Flux} & \colhead{Class} & \colhead{lat.} & \colhead{long.} & \colhead{Index}} 
408: \startdata
409: 8073 & 08.17.1997 & 229.25    & 129897   & $\alpha$ & 14.5 & 282.0 & 40553\\
410: 8243 & 06.18.1998 & 694.59    & 455180   & $\beta$  & 18.3 & 214.5 & 47871\\
411: 8397 & 12.04.1998 & 745.76    & 492968   & $\beta$  & 16.5 & 140.6 & 51932\\
412: 8402 & 12.05.1998 & 298.17    & 187218   & $\beta$  & 17.0 & 122.5 & 51958\\
413: 8402 & 12.07.1998 & 788.06    & 436916   & $\beta$  & 16.8 & 121.8 & 51983\\
414: 8602 & 06.29.1999 & 867.21    & 543457   & $\beta$  & 19.1 & 298.7 & 56893\\ 
415: 8742 & 10.30.1999 & 796.93    & 527631   & $\beta$  &  7.0 & 121.0 & 59829\\
416: 8760 & 11.10.1999 & 1677.79   & 1019900  & $\beta$  & 13.0 & 330.0 & 60102\\
417: 9236 & 11.24.2000 & 1849.05   & 1062150  & $\beta$  & 18.5 & 359.0 & 69216\\
418: 9493 & 06.12.2001 & 311.13    & 202267   & $\beta$  & 5.8  & 235.2 & 74032\\
419: 0061 & 08.09.2002 & 576.55    & 363652   & $\beta - \delta$ & 8.5 & 45.2 & 84183\\
420: 0330 & 04.09.2003 & 1288.19   & 819593   & $\beta - \gamma$ & 6.2 & 82.2  & 90015\\
421: 0387 & 06.23.2003 & 617.49    & 335377   & $\beta - \gamma$ & 17.0 & 166.5 & 91806\\
422: 0615 & 05.21.2004 & 404.61    & 216729   & $\alpha$ & 15.6 & 93.7 & 99812\\
423: 0689 & 10.27.2004 & 206.06    & 114991   & $\alpha$ & 9.7 & 155.0 & 103621\\
424: 0898 & 07.03.2006 & 1194.03   & 699494   & $\beta$  &-5.5 & 329.0 & 118363\\
425: \enddata
426: \tablenotetext{a}{NOAA is the active region number hosting the sunspot, as provided by the NOAA, the Area is in Mm$^2$, Flux is the l.o.s magnetic field flux in G Mm$^2$, \mbox{lat.} is the Carrington latitude and \mbox{long.} is the Carrington longitude of the center of the datacube where the active region is located, and Index is the MDI index of the first frame of the datacube retrieved on the SOI/MDI website. NOAA number and magnetic classes come from the website: http://www.solar.ifa.hawaii.edu/ARMaps/}
427: \end{deluxetable}
428: 
429: \clearpage
430: 
431: \begin{figure}
432: \plotone{f1.ps}
433: \caption{MDI continuum intensity of the 16 solar active regions studied in this paper, in the order listed in Table \ref{table.spot} (from the upper left to the lower right panel). The boundaries of the umbra and penumbra are overplotted: they correspond to $57\%$ and $85\%$ of the continuum intensity, respectively. The greyscale is the same for every panel.\label{f1}}
434: \end{figure}
435: 
436: \clearpage
437: 
438: \begin{figure}
439: \plotone{f2.ps}
440: \caption{Mean travel-time perturbation maps $\delta \tau_{\mathrm{mean}}(\mathbf{r},\Delta)$ for the distance $\Delta=11.6$ Mm and for the 16 sunspots studied. The greyscale is the same for every panel and was truncated to emphasize the dark rings. The boundaries of the umbra and penumbra, as defined on Figure \ref{f1}, are overplotted.\label{f2}}
441: \end{figure}
442: 
443: \clearpage
444: 
445: \begin{figure}
446: \epsscale{0.7}
447: \plotone{f3.ps}
448: \caption{Azimuthal average of $\delta \tau_{\mathrm{mean}}(\mathbf{r},\Delta)$ for $\Delta=6.2$ Mm  (upper panel), $\Delta=8.7$ Mm (second panel from the top), and $\Delta=11.6$ Mm (third panel) and for 5 sunspots: NOAA 9493 (solid line), 8073 (short-dashed line), 0615 (dot-dashed line), 8402 (dot-dot-dot-dashed line), and 0689 (long-dashed line). The lower panel shows an azimuthal average of the normalized continuum intensity.}
449: \label{f9}
450: \end{figure}
451: 
452: \clearpage
453: 
454: \begin{figure}
455: \epsscale{1.0}
456: \plotone{f4.ps}
457: \caption{Vertical cut in the inverted sound-speed perturbation for the active region NOAA 8397.\label{f6}}
458: \end{figure}
459: 
460: \clearpage
461: 
462: \begin{figure}
463: \plotone{f5.ps}
464: \caption{Inversion results $\delta c^2/c^2$: we show a horizontal cut at a depth of $z=-2.38$ to $z=-1.42$ Mm. The greyscale is the same for every panel and was truncated to emphasize the bright rings. The boundaries of the umbra and penumbra, as defined on Figure \ref{f1}, are overplotted.\label{f5}}
465: \end{figure}
466: 
467: \clearpage
468: 
469: \begin{figure}
470: \plotone{f6.ps}
471: \caption{Inversion results $\delta c^2/c^2$: we show a horizontal cut at a depth of $z=-1.42$ to $z=-0.62$ Mm. The greyscale is the same for every panel and was truncated to emphasize the bright rings. The boundaries of the umbra and penumbra, as defined on Figure \ref{f1}, are overplotted.\label{f8}}
472: \end{figure}
473: 
474: \clearpage
475: 
476: \begin{figure}
477: \epsscale{.8}
478: \plotone{f7.ps}
479: \caption{Azimuthal averages of the inverted sound-speed perturbation at a depth of $z=-2.38$ to $z=-1.42$ Mm (upper panel) and $z=-1.42$ to $z=-0.62$ Mm (lower panel), as a function of the radial distance to the spot center, for 5 sunspots: ARs 9493 (solid line), 8073 (short-dashed line), 0615 (dot-dashed line), 8402 (dot-dot-dot-dashed line), and 0689 (long-dashed line).\label{f4}}
480: \end{figure}
481: 
482: \clearpage
483: 
484: \begin{figure}
485: \epsscale{.8}
486: \plotone{f8.ps}
487: \caption{Upper panel: azimuthal average of $\delta \tau_{\mathrm{mean}}(\mathbf{r},\Delta)$ for the distance $\Delta=11.6$ Mm and the active Region NOAA 8397. The solid line is before the power correction of Rajaguru et \mbox{al.} (2006), the dashed line is after correction. Lower panel: same plot for the active region NOAA 8073.\label{f11}}
488: \end{figure}
489: 
490: \clearpage
491: 
492: \begin{figure}
493: \epsscale{1.0}
494: \plotone{f9.ps}
495: \caption{Travel-time maps $\delta \tau_{\mathrm{mean}}(\mathbf{r},\Delta)$ for the distance $\Delta=11.6$ Mm and the active Region NOAA 9493. Left panel: with a frequency filter centered on 3 mHz; central panel: filter centered on 4 mHz; right panel: filter centered on $4.5$ mHz. The greyscale is the same for all three panels.\label{f10}}
496: \end{figure}
497: 
498: \clearpage
499: 
500: \begin{figure}
501: \epsscale{.8}
502: \plotone{f10.ps}
503: \caption{Azimuthal average of $\delta \tau_{\mathrm{o}}(\mathbf{r},\Delta)$ (outgoing travel times; upper panel) and $\delta \tau_{\mathrm{i}}(\mathbf{r},\Delta)$ (ingoing travel times; lower panel) for $\Delta=11.6$ Mm and for 5 sunspots: NOAA 9493 (solid line), 8073 (short-dashed line), 0615 (dot-dashed line), 8402 (dot-dot-dot-dashed line), and 0689 (long-dashed line).\label{f13}}
504: \end{figure}
505: 
506: \clearpage
507: 
508: \begin{figure}
509: \epsscale{1.0}
510: \plotone{f11.ps}
511: \caption{Maximum value of the sound-speed perturbations in the rings surrounding the spots for the layer $z=-2.38$ to $z=-1.42$ Mm, as a function of the sunspot magnetic flux (see Table 1). The error bars are $\pm$ the rms variation of the inverted sound-speed perturbations at the selected layer. The solid line is the result of a linear fit.\label{f12}}
512: \end{figure}
513: 
514: \end{document}
515: