A&A 444, 245-255 (2005)
M. Leitzinger 1 - P. N. Brandt 2 - A. Hanslmeier 1 - W. Pötzi 1 - J. Hirzberger 1
1 - Institut für Physik, IGAM, Karl-Franzens Universität Graz, Universitätsplatz 5, 8010 Graz, Austria
2 - Kiepenheuer-Institut für Sonnenphysik, Schöneckstr. 6, 79104 Freiburg, Germany
Received 29 March 2005 / Accepted 3 June 2005
Using a 45.5-h time series of photospheric flow fields generated from a set of high-resolution continuum images (SOHO/MDI) we analyze the dynamics of solar mesogranule features. The series was prepared applying a local correlation tracking algorithm with a 4.8 FWHM window. By computing 1-h running means in time steps of 10 min we generate 267 averaged divergence maps that are segmented to obtain binary maps. A tracking algorithm determines lifetimes and barycenter coordinates of regions of positive divergence defined as mesogranules (MGs). If we analyze features of lifetimes 1 h and of areas 5 Mm2 we find a mean drift velocity of 304 m s-1 (with variation of 180 m s-1), a mean travel distance of Mm, a mean lifetime of h, and a 1/e decay time of 1.6 h for a total of 2022 MGs. The advective motion of MGs within supergranules is seen for 50 to 70% of the long-lived (4 h) MGs while the short-lived ones move irregularly. If only the long-lived MGs are further analyzed the drift velocities reduce to 207 m s-1 and the travel distances increase to 4.1 Mm on average, which is an appreciable fraction of the supergranular radius. The results are largely independent of the divergence segmentation level.
Key words: convection - Sun: granulation - Sun: photosphere
In 1981, November et al. found a "fairly stationary pattern of cellular flow'' from time averages of Doppler velocities measured at disk center. The observed pattern had a spatial scale of 5-10 Mm, a vertical rms velocity amplitude of about 60 m s-1and a lifetime of at least 2 h. They termed this pattern mesogranulation (MG) and speculated that it was the possible "missing'' scale between granulation and supergranulation (November et al. 1981).
The question of the origin of MG is still debated. A pattern of so-called active (i.e., repeatedly splitting) granules was found by Oda (1984); its mesh size was comparable to mesogranules. Also Koutchmy & Lebecq (1986) confirmed the existence of quasi-stationary cells of sizes in the range of 10-15 . Dialetis et al. (1988) investigated long-lived granules and found that they are not randomly distributed but that their location formed a meso-scale mesh. A similar pattern with a mean cell size of 7 was seen by Muller et al. (1990); it was formed by large as well as small granules. In an investigation of granular properties on a mesogranular scale Brandt et al. (1991) showed that properties like area, lifetime, brightness and growth rate vary according to where the corresponding granules are located in the MG. Hirzberger et al. (1999) investigated a time series of solar granulation images. They found that small granules (also granules of short lifetimes) show a slight tendency to be located in downflow regions and larger granules in upflow regions of the MG. A connection of active granules and the mesogranular flow was also suggested by Müller et al. (2001). They compared the mean distance of long-lived active granules and the mesogranular scale. Also Pötzi et al. (2003) found that granules are not distributed at random, instead "fragmenting, ring-like, and merging'' granules are preferentially located in upflow regions (divergence) and "fading'' granules in downflow (convergence) regions of the MG.
The question: "Is MG a scale of convection that is distinct from granulation and supergranulation?'', is treated by many authors. Using power, phase, and coherence spectra derived from a time series of high spatial resolution spectra of photospheric Fraunhofer lines, Deubner (1989) studied the physical properties of the MG (see Table 1). He reports a "genuine convective flow'' in his results. Wang (1989) used long sequences of Dopplergrams and found a typical scale of 7 Mm, but no sign of the cellular convection. In an analysis of power spectra, derived from a series of white light images and monochromatic filter scans in two Fraunhofer lines, Straus & Bonaccini (1997) did not find a "separate regime of mesogranulation distinct from granulation''. Ginet & Simon (1992), Straus et al. (1992) and Hathaway et al. (2000) also find no evidence for a separate mesogranular regime from various analyzing methods, whereas Chou et al. (1991), Bachmann et al. (1997), Ueno & Kitai (1998) and Domínguez Cerdeña (2003) came to the conclusion that there is a convective regime at scales just below supergranulation.
Rieutord et al. (2000) found that "mesogranulation is not a true scale of solar convection but the combination of the effects of both highly energetic granules, which give birth to strong positive divergences'' ... "and averaging effects of data processing''. From numerical experiments, Cattaneo et al. (2001) concluded that mesogranules owe their origin to the interaction between granules, while Steiner (2003), also from numerical simulations, found recurrently fragmenting granules that drive a horizontal flow field of mesogranular scale. Similarly, Rast (2003) stated that a simple n-body simulation of the interaction of granular downflow plumes was successful at reproducing the observed spatial and temporal scales of both MG and supergranules. Roudier et al. (2003) confirm the suggestions of Straus & Bonaccini (1997) and Rieutord et al. (2000), that MG is not a "specific scale of convection'' ... "but just the large-scale extension of granulation''. From a two-dimensional analysis of the granular intensity field they found that a "significant fraction of granules'' form so-called trees of fragmenting granules, which are very long-lived, a result that was previously found from a one-dimensional time-slice analysis by Müller et al. (2001).
The advective motion of MGs towards the borders of the supergranulation was first described by Muller et al. (1992). Also DeRosa & Toomre (1998, 2004) found convincing evidence in their results from SOHO/MDI Dopplergrams that MGs are advected by the supergranulation. Recently, Shine et al. (2000, henceforth called SSH) analyzed a 45.5-h series of high-resolution continuum images from SOHO/MDI to investigate supergranule and mesogranule evolution. Three statements in the summary of their paper, which are listed below, were the motivation for the present work:
The basis of the present work is a time series of 2588 flow fields with a time separation of one minute. These flow fields result from applying a LCT (Local Correlation Tracking) algorithm to a time series of continuum images, obtained by the SOHO/MDI instrument from 17th to 18th of January 1997 (original image size: pix, where 1 pix = 0.6 ). After projection onto a latitude-longitude grid the finally extracted images, that follow the solar rotation, have a size of MDI pixels. For the LCT algorithm a FWHM of 8 pix was used, while the cell centers were spaced 4 pix apart, thus oversampling by a factor of 2. This cell spacing yields flow fields of cells with a spatial resolution of 4.8 or 3.5 Mm. Because of several gaps in the data set, which add up to about 2.5 h (141 missing flow fields), we had to interpolate the missing flow fields. Since most of the gaps were short, i.e., 38 out of 49 consisted of 2 missing flow fields and only one was 11 flow fields long, we used a simple interpolation method: the missing flow fields were replaced by the mean of the 30 flow fields before and the 30 flow fields after the gap. Further we rebinned the original flow fields by a factor of 3 in each direction (new field: pix of 1.6 or 1.16 Mm size each) to obtain a smoother representation of the flow vectors. Next, we computed the divergence of the flow fields and 1-h running means with a time separation of 10 min (for an example see Fig. 1).
Table 1: Values of mesogranule lifetime and drift velocity from different authors.
|Figure 1: First one hour average divergence map; bright areas: strong divergence (upflow), dark areas: strong convergence (downflow).|
|Open with DEXTER|
To define mesogranular features we generated binary images from the divergence maps; we call them segmented maps. For the segmentation we used a threshold method, and defined every pixel above the threshold as a mesogranular pixel - in the following we call these contiguous positive divergence areas mesogranules.
four levels of the cumulative histogram of the 267 1-h average
divergence images as thresholds, i.e., 80%, 85%, 90% and 95% of the maximum divergence value
(Fig. 2 shows the 80% and the 90% level).
|Figure 2: Mean cumulative histogram of all divergence values of the 267 one-hour averages. The vertical lines show the divergence thresholds at 80% and 90%. The values of divergence above the threshold are defined as mesogranular pixels.|
|Open with DEXTER|
Areas of Mm2 of divergence images and segmented maps are shown in Fig. 3 to demonstrate the influence of the segmentation level on the segmented features. The areas show two main differences between higher and lower segmentation levels: with increasing level both the size and the number of the segmented features decreases. The last step in the preparation process was the identification of the features in order to track them automatically. In each map each feature was assigned its own identification number.
|Figure 3: Areas of Mm2 of the original divergence (bright: positive) and segmented maps with a cadence of 10 min. The first row shows three successive cutouts of divergence maps. The second to fifth rows show the corresponding segmented maps for segmentation levels 80% to 95%. The influence of the segmentation level on the segmented features, i.e., decreasing area and number of features with increasing level, is clearly seen.|
|Open with DEXTER|
The feature tracking algorithm used in this paper is based upon the algorithm of Hirzberger et al. (1999) which was originally developed for the tracking of granules. The algorithm tracks forward and backward in time from a given starting binary map. We chose every fifth segmented map (i.e., time gaps of 50 min) as starting point, in order to consider also short-lived features. The adaptation of the algorithm to its present task mainly concerned the interrupt conditions, that select correct followers or predecessors according to the area and the barycenter of the tracked feature; the interrupt conditions are:
Table 2: Mean values of the mesogranular characteristics as function of four segmentation levels; most values are given together with their scatter around the mean.
The evolution time of the corks is 5 h, sufficient time to advect them to the supergranular boundaries. On the other hand, as one can see from Fig. 7, the lifetimes of MGs vary from 1 h to >10 h. Because of this difference in time, the vectors are overplotted on cork maps corresponding to the middle age of the vectors. In some cases, where the lifetime of mesogranules is very long, it can happen that single long-lived mesogranules (or vectors, respectively) cross the supergranular boundaries - possibly because of the supergranular evolution over this long time span. However, except for these very rare cases, the maps in Figs. 4 and 5 show that no mesogranular feature crosses the supergranular boundary. Both the long-lived (marked in red) and the far traveling MGs perform a nearly radial advective motion towards the edges of the supergranules; however, there are some exceptions to this behavior - mainly in cases of small or not well defined supergranules. On the other hand, many short-lived features move rather irregularly within the supergranulation and never reach its boundary.
|Figure 4: Four plots of mesogranule trajectories (cadence of 5 h) overlayed with cork maps (running time of the corks is 5 h). Only mesogranules with a lifetime of 1 h were used. Blue-colored vectors correspond to mesogranules with a lifetime <4 h, whereas red-colored vectors correspond to mesogranules with a lifetime 4 h.|
|Open with DEXTER|
|Figure 5: Cork and vector plots (cadence 2.5 h) spanning an area of Mm2 including one big and several smaller supergranules. The running time of the corks is the same as in Fig. 4 (5 h). Red-colored vectors correspond to mesogranules with a lifetime of 4 h, the blue-colored vectors to mesogranules with a lifetime <4 h. The time sequence begins in the upper left corner with a cutout generated from starting image No. 30, proceeding line by line to the lower right corner (starting image No. 150).|
|Open with DEXTER|
However, these values are of limited interest since they depend strongly on the segmentation level (cf. Table 2) while the other mesogranular characteristics do not. The area histogram shows a steep decline from the smallest features that were analyzed (i.e., 5 Mm2) to values >30 Mm2. The mean travel distance of the MGs is 2.5 Mm (1.8) with a median of 2.1 Mm. Here most of the MGs obviously travel less than 2 to 3 Mm which is only a small fraction of the supergranular radius while only 8.5% reach distances of 5 Mm, i.e., values comparable to the supergranular radius.
Table 2 shows convincingly that, while the number of tracked MGs drops by a factor of 2.3 and their areas decrease by a factor of 2.1 with increasing segmentation level, the values of drift velocity, lifetime, decay time (except at the lowest level) and travel distance vary only slightly more than . Moreover, only the drift velocity exhibits a downward trend whereas both lifetime and travel distance go through a flat maximum with increasing segmentation level. This finding gives us confidence in the results - although, of course, they depend on the specific parameters (1 h averaging time and 4.8 LCT window) used to derive the horizontal velocities.
|Figure 6: Histogram of MG drift velocities for the segmentation level of 90%. Only mesogranules of lifetimes 1 h and of areas 5 Mm2 were used.|
|Open with DEXTER|
|Figure 7: Number of mesogranules as function of lifetime (in log-lin scale) for the segmentation level of 90%. Only mesogranules of lifetimes 1 h and of areas 5 were used. The exponential fits for the four segmentation levels are overplotted in different linestyles.|
|Open with DEXTER|
|Figure 8: Histogram of MG areas for the segmentation level of 90%. Only mesogranules of lifetimes 1 h and of areas 5 were used.|
|Open with DEXTER|
|Figure 9: Histogram of MG travel distances for the segmentation level of 90%. Only mesogranules of lifetimes 1 h and of areas 5 Mm2 were used.|
|Open with DEXTER|
|Figure 10: Scatter plot of MG lifetime vs. drift velocity for the segmentation level of 90%. Triangle symbols mark mean values of velocity for the lifetime bins.|
|Open with DEXTER|
|Figure 11: Scatter plot of MG lifetime vs. travel distance for the segmentation level of 90%. Triangle symbols mark mean values of travel distance for the lifetime bins.|
|Open with DEXTER|
|Figure 12: Scatter plot of MG lifetime vs. area for the segmentation level of 90%. Triangle symbols mark mean values of area for the lifetime bins.|
|Open with DEXTER|
|Figure 13: Scatter plot of MG travel distance vs. area for the segmentation level of 90%.|
|Open with DEXTER|
The first scatter plot (Fig. 10) reveals a clear tendency of the drift velocity to decrease with increasing lifetime. If we introduce a limit of 4 h in order to distinguish between short-lived and long-lived MGs, we find that the short-lived ones drift at velocities between 0.25 and 0.4 km s-1 on average, showing a very strong variability between 0 and 1 km s-1, while the long-lived ones drift at about 0.1 to 0.3 km s-1with much smaller variability. Concerning the travel distance, Fig. 11 indicates values of 1 to 3 Mm for the short-lived MGs and a wide range from 3 to more than 10 Mm for the long-lived ones, in other words a clear increase of travel distance with lifetime. The relation between MG area and lifetime (Fig. 12) is characterized by an increase of lifetime with area: the short-lived MGs show areas between 8 and 10 Mm2 while the areas of the long-lived ones range from 8 to >18 Mm2. Finally, the relation between MG travel distance and area reveals no specific trend; most of the MGs have small areas (5-10 Mm2) and travel 0-3 Mm - a result which could already be read from the histograms (Figs. 8 and 9).
In the following we discuss our results with reference to those quoted by SSH and their relation to some previous measurements of MG characteristics. When comparing these results, each of the quoted results depends in a characteristic way on the specific parameters chosen for the analysis, especially on the spatial and temporal filter that was applied to separate granular from mesogranular properties - the main reason being that the velocity power spectrum shows a continuous decrease from supergranular down to granular spatial scales, as Hathaway et al. (2000) have demonstrated.
One would expect that the advective flow velocities of the MGs, which are measured here at the photospheric level, are closely related to the horizontal velocities measured in supergranules by spectroscopic methods; for these Simon (2001) quotes a range from 200-500 m s-1, which is compatible with the range of values of 200-300 m s-1 (Table 3) found here - especially if one considers the wide spread ( 100-180 m s-1) of the scatter around the mean. Simon (1967) found "...that the granules tend to move toward the network boundaries, apparently drifting in the direction of the supergranule flow pattern''. In the present work we confirm the same pattern for the divergence of this granule drift.
Table 3: Mean characteristics of MGs as function of the lower limit of lifetime; each mean value is given together with its scatter around the mean.
In Fig. 5 of their publication SSH show the distribution of the flow velocities derived from their LCT analysis of granule conglomerates: it has a peak around 450 m s-1 and shows a broad tail reaching beyond 1000 m s-1. In an analysis of a high-resolution granulation series observed at Pic du Midi Roudier et al. (1999) found a velocity histogram that peaks around 0.6 km s-1 and reaches up to 2.5 km s-1. The difference between the findings of SSH and Roudier et al. may be explained by the large difference in the FWHM of the LCT windows: while Roudier et al. used a window of 0.7 and thus were able to track single granules, SSH's window size was 4.8 and was, therefore, only capable of tracking groups of granules.
Tracer particles (corks) traveling at, e.g., 0.5 km s-1 for 4-6 h cover the supergranular radius; therefore the statement by SSH is very appropriate: "Corks (tracers) are cleared from the interiors of medium (20 Mm) supergranules in 4 h, and from large ones (30 Mm) in 6 h''. Our Figs. 4 and 5, produced with cork travel times of 5 h, confirm this, except in regions where magnetic fields are present, as is the case in Fig. 4 in a broader region around x=70 Mm, y=70 Mm. However, the motion of the corks represented by the flow fields is not necessarily the same thing as the motion of its derivative, the divergence, which by definition represents the MG. This is consistent with our finding of MG drift velocities which are about a factor of 2 lower than the typical cork motion, as determined from tracking granules or groups of granules.
From our re-analysis of a 45.5-h set of flow fields that were derived by Shine et al. (2000) by tracking groups of granules on SOHO/MDI continuum images we find that 50 to 70% of the MGs of lifetimes 4 h exhibit a more or less radial motion within supergranules; shorter lived mesogranules travel in random directions. We find a wide range of sizes, lifetimes, travel distances and velocities in our sample of mesogranules. If the analysis is limited to MGs of lifetimes 1 h and of areas 5 Mm2 we find the following mean values: lifetime 158 (107) min, travel distance 2.5 (1.8) Mm, and drift velocity 304 (180) m s-1. A novel result is the 1/e decay time of 1.6 h found here. If the analysis is limited to MGs of lifetimes 4 h the resulting travel distances increase to 4.1 (2.3) Mm while the corresponding drift velocities reduce to 207 (111) m s-1. The travel distances thus are 1/3 to 1/2 of the commonly accepted values of the supergranular radius of 6-10 Mm (cf. Hagenaar et al. 1997; DeRosa & Toomre 2004) and the drift velocities are comparable to the horizontal flow velocity of the supergranulation as determined spectroscopically (Simon 2001); the latter fact suggests the interpretation that the MG elements are carried along with the horizontal supergranular flow. In their recent careful investigation of supergranule evolution, DeRosa & Toomre (2004) tracked by LCT the proper motion of meso-scale elements after high-pass filtering MDI Dopplergrams and thus determined the supergranular pattern from the horizontal outflows observed.
The differences between our results and those of SSH mainly may be due to the fact that their visual inspection of movie sequences and space-time diagrams may have preferred other sub-sets of features and thus have been more prone to selection effects than the more objective method applied here. The lifetimes found by other authors (cf. Table 1) range from 30 min to 3 h and their drift velocities are in the range 300-500 m s-1 which is in good agreement with our values, especially when one keeps in mind that due to the specific method applied here (large LCT window of 4.8 , 1-h running means) the slowly evolving large-scale component of the mesogranulation was studied. On the other hand, the large field of view and the long duration of 45.5 h with a 1-min cadence of the observational data give us strong confidence in the statistical stability and significance of the results.
We thank R. A. Shine for supplying the pre-reduced data and M. Stix as well as O. Steiner for helpful comments on the manuscript.
Divergence (Zip file 27.6MB)
Segmentation (Zip file 3.4MB)