08 APRIL 2005  Total Solar Eclipse
21d 18' 06'' S, 128d 22' 31"W

Glenn Schneider, Steward Observatory, University of Arizona

In collaboration with Jay Pasachoff, Williams College

With technical, programmatic, and personal support and contributions (thanks, guys!) from:

Joel Moskowitz, Jay Friedland, Jean-Luc Dighaye, Michael Gill, John Beattie, Craig Small, Robyn Small, Charles Cooper, William Whiddon (postumously), and Steve Kolodny (unfortunately, in abstencia).

And, a special acknowledgement to Captain Gilles Boussard of the M/S Paul Gauguin for his all of his efforts.


A coronal imaging program, with instrumentation similar to that used for the 23 November 2003 total solar eclipse observed from 35,000 ft over the Antarctic, was planned for the 32s of totality observed on 8 April 2005 from the South Pacific aboard the M/S Paul Gauguin.  Despite our best efforts, however, the weather had other ideas and we were unsuccessful in obtaining the planned imagery with gyro stabilized cameras. This was due (primarily) to clouds obscuring the Sun until – literally – the advent of totality. Still and video images were obtained (e.g., by Jay Friedland and Joel Moskowitz), but this situation precluded properly balancing and detorquing a 3-axis gyro stabilized instrument platform for the required Sun-pointing because the Sun could not be acquired before second contact. As a result, the ship's motion (to dodge clouds) induced wind-loads on the instrument platform exceeding the compliance of the gyro control stabilizer units with the platform not properly balanced.

Assessing the cloud/wind conditions at ~ CII - 90 minutes.
Pre-eclipse planning on the bridge.

Gyro platform with 2 film and video cameras showing.

Gyro platform with chilled-water cooled SBIG CCD camera showing.
 
Miraculously, the offending clouds began to part with the advent of the second contact diamond ring, shown in the left/right stereogram below (composed from frame-extracted video imagery obtained by Joel Moskowitz). The disappearance of the last remnant of the solar photosphere was virtually coincidental with the disappearance of the clouds along the line-of-sight to the sun, too late to permit stable image acquisition with the cameras on the gyro platform shown above, but affording a stunning and  and spectacular view of totality.


(For information on viewing left/right stereograms CLICK HERE).

Despite the clouds only seconds earlier, imagery of totality with high information content and dynamic range in the region of the inner to mid corona was independently obtained by William Whiddon, a fellow passenger on the M/S Paul Gauguin. Bill very kindly had provided me a with a subset of his images, which are quite spectacular even in raw form.  This imagery, when suitably processed, as discussed here, specifically informs on complex inner coronal structures which were seen. In mid-September 2005, Bill had offered to supply me with the full set of his raw images for additional processing and subsequent analysis.


In Memory of William Whiddon, TRW and founding member of SBAS

Very sadly, and unexpectedly,  I am sorry to say that soon after Bill's kind offer, I learned from Nina, Bill’s wife, that Bill died on 11 October 2005 due to complications arising from acute leukemia.  This was, to say the very least, quite a shock. Though I had only met Bill for the first time during our recent travels to the South Pacific, it was quite clear to me that his passion for solar eclipses was very deep indeed.  I had looked forward, following our too-brief post-eclipse communications, to fostering a collegial relationship which would have brought us together in future eclipses – which of course will now never happen.  Nina had passed along to me that “he was so thrilled that you guys {myself and Alex Filippenko} liked his South Pacific eclipse pictures.  Other people had told them they were quite impressed...” My admiration for his work was very justly deserved, and I hope I had really adequately expressed that to him. 


The William Whiddon TSE 2005 Gallery of Eclipse Images


Bill's "fisheye" view of Totality seconds after an offending cloud (left of the Sun) which obscured the Sun before second contact cleared away.
CLICK ON ANY IMAGE TO SEE AT TWICE THE SIZE

Bill had noted that the above images "were taken with a Nikon D2X with a Nikkor 70-200mm f/2.8G ED IF VR, coupled with a TC-20EII doubler.  The vibration reduction mode was not used.  The images were shot at 600mm EFL (due to the 1.5x scale factor of this 12 megapixel camera).  The camera and lens were mounted on a Wemberley single-tine fork arm, balanced for the elevation of totality, and with the az and elclutches loosened.  I worked out a system with a precision az/el adjuster in '98 that enabled me to take 1/2 sec exposures from aboard ship by adjusting the pointing to the extremum of ship motion.  However, I knew that system would not work on this ocean (with waves from multiple directions and swells from either the Gulf of Alaska or the Southern Ocean), or with the smaller ship I was on.  And with only about 31 seconds of totality (as opposed to 3m 42s in '98), it was essential to be able to acquire and point quickly.  So to meet the needs of this eclipse, I rested my hands on the tripod head, and by moving the balanced assembly carefully with my fingers I was able to guide the camera to a predict-ahead position for the extremum of ship motion, and fire a sequence of 7 exposures auto-bracketed from 1/1000 through 1/15 seconds.  The camera was hand-fired (no point in a cable release when bear-hugging it).  At 7 frames per second firing rate, it only took 1.5 or 2 seconds to do the sequence.  In this manner I was able to get in some 66 exposures in the brief totality -  with time to fire another camera with a 10.2mm fisheye, and even time to do a little visual observing.  Of course, I did all the usual shipboard tricks of finding a vibration-free, wind-protected portion of the deck to set up on, suppressing vibrations in the tripod legs, etc.

To acquire the sun, I used a TeleVue Sol Searcher mounted to the Wemberley on a flash bracket and boresighted with the lens.  With this system and technique, I was able to acquire the sun in as little as 2 or 3 seconds, enabling me to capture the first diamond ring and the surrounding diffractive corona aureole immediately after cloudbreak. 

The exposure for the chromocircle shot {bottom left} was I believe 1/1000 sec, and that for the early totality shot was I believe 1/125 {top, right}.  All exposures were at ISO 200. 

I was very happy with my setup, as it proved very robust and adaptable for shipboard eclipses.  Setup and teardown were quick, enabling me to outrun a sudden freshet after C3.  And since exposures were cheap with the digital camera, I was able to make over 1000 test exposures aboard ship to work out the bugs. "

Additional imagery of TSE 2005 from the M/S Paul Gauguin acquired by William Whiddon may be found HERE.


DIGGING INTO THE "DIRT" (more digital magic)

The image in the upper-left panel above image actually contains a significant amount of structural and morphological information in the inner coronal region which might not be apparent on casual viewing. This may be "brought out" with suitable image enhancement (i.e., quantitative digital image processing).  To illustrate this point, the pair of images below (rotated to put North "up") was processed specifically to bring out the larger spatial scale structures in the inner corona.




Left: Pre-Processed Image, Right: Post-Processed.


In a bit more detail, the image processing was done in manner specifically to improve the visibility of those features in the inner coronal which:

(a) dominantly have more power at mid and low spatial frequencies and are correlated in two dimensions on spatial scales exceeding an estimate of a few resolution elements while simultaneously,

(b) suppressing high spatial frequency information which is uncorrelated in the azimuthal direction on the same spatial scales. 

The method developed and employed to do this works purely in the image domain. It is a multi-pass spatial filtering technique and does NOT employ image deconvolution (blind, adaptive, or using an optically/empirically derived PSF template). 

Actually, you can see (*rather dramatically*) the nature of the enhancement in this spatial sampling domain by "BLINKING" THE IMAGES.


Unwrapping the Chromosphere and Innermost Corona.


Inner Corona Along the Solar Limb: Heliocentric Polar -> Cartesion Projection

CLICK THE IMAGE TO SEE AT TWICE THE SCALE.


CLICK THE IMAGE TO SEE AT TWICE THE SCALE.
ABOVE: The post-processed image has been "unwrapped" (so the photospheric gravity gradient is "downward") to show the larger  spatial scale structures in the inner regions of the corona. 

BELOW: The three images shown in the Whiddon Gallery, exclusive of the diamond ring image {lower, right}, were combined see better the chromospheric region adjoining the very innermost part of the corona.  The details of that image combination process are more conventional then the filtering method described below. Saturated pixels are masked in individual frames, linearity transformations are applied to non-saturated pixels before combining to compensate for different exposure times and responsivities.

In both images, left corresponds to a position angle of zero (i.e. "North up" in the coronal image) with a the "clockwise" direction increasing to the right.


The Nitty-Gritty of Spatial Filtering.

The spatial filtering of the above coronal imagery (done in the the image domain) has a direct analog in the Fourier (frequency) domain that would be akin to using an inverse transformation (from frequency back to image space) with elliptically apodized truncation of the 2D Fourier components.  The complexity of the latter approach would not be in the algorithm, but rather "knowing" WHERE to (optimally) truncate the transformation (i.e., the bi-axial extent of the truncation ellipse in the Fourier domain), as well as how to apodize it.  This is solvable for a given desired dominant spatial frequency extraction - but becomes complex in the presence of "real world" noise. 

Alternate approaches "recover" information in different spatial domains with varying degrees of efficacy, that depends in part on the local signal-to-noise (S/N) ratio in the image. The very successful processing done by Drückmuller on their input images (in their approach) works very well, as most recently demonstrated with intrinsically high S/N input imagery from TSE 2005 (a median combination of 32 images, improving the S/N over a single image by a factor of 5.7, and critically important for recovering information at higher spatial frequencies in regions of low surface brightness).


Yet, Another Approach

The approach taken here applies iterative/re-entrant (multi-pass), and unequally-weighted, subtractions of azimuthally-restricted heliocentered rotational medians of successive input/output images.  At each iteration step azimuthal "integrations" are performed on the re-entered image.  These, actually, are not integrations, but numerical medians.  For every azimuth angle, in radial "zones" resampled to twice the original pixel scale with sinc-function apodized bicubic interpolation to half an original pixel, an azimuthal median is computed over a restricted range of azimuth angles about every pixel.  That angular range is +/- twice the "size" of the largest spatial features to be recovered  to a limit of, typically, two resolution elements ifthe image is Nyquist sampled. 

The iterations begin with the largest spatial scales to be recovered azimuthally, and becoming more restrictive (filtering smaller "features" on successive iterations) guided by the per-pixel S/N in each radial zone.  The rotationally-medianed image is then subtracted from the input image after appropriate weighing, so no negative residuals arise on the global (360 degree) azimuthal average at all radii after subtraction.  This process, in some sense, is loosely akin to damped radial gradient filtering in combination with unsharp masking, but avoids the pitfalls of artifact creation in both of those methods due to over-aggressive radial and local background gradient subtractions.

This method implicitly results in an image which "intelligently" collapses the azimuthally-global radial dynamic range (as "near-perfect" radial gradient filtering would do).  More importantly, it rejects (filters) features which have azimuthal spatial scales comparable to those of the sampling noise if (AND ONLY IF) they are correlated over multiple pixel scales (i.e., many resolution elements) in the radial direction, but not so if they are correlated over similar scales in the azimuthal direction.  I.e., the filtering has a polar coordinate spatial dependence.  Hence, features like very fine strands of the polar "brushes" are removed (the "there is no free lunch" theorem of image processing) - but "fatter" features like coronal loops (e.g., at the "one o'clock position") and the noted 3-o'clock feature remain - with very enhanced visibility.

The process is illustrated in the panel to the left progressing through four iteration from the input image input (top) to "final "(bottom).  Click HERE or on the images to the left, to view these itrative enhancements to the input image at higher spatial resolution.



Digression: (Philosophical) Comments on Image Deconvolution.

Deconvolution methods have a (not undeserved) reputation for introducing image artifacts on many spatial scales depending upon the nature of the deconvolution, the structure and fidelity of the PSF kernel, and the (local) S/N in the image to be deconvolved.  With sufficient S/N, the highly successful adaptive kernel deconvolution method of Drückmuller can, to a significant degree, avoid that. The loss of some photometric efficacy, at the expense of maintaining/recovering astrometric (i.e., morphological) information (this is affectionately known as the "no free lunch" theorem of signal processing which I am sure Heisenburg would have approved of {with deconolution - loosely speaking - position and flux density act as conjugate variables}).

The burden of proof put upon investigators to demonstrate the efficacy of a deconvolution method, as applied to specific data sets (by other skeptical researchers and journal referees), they are restoring or enhancing, is not misplaced.  Many, historical, examples speak to the need for that (e.g., see "The Restoration of HST Images and Spectra"; Proceedings of a Workshop held at the Space Telescope Science Institute in Baltimore, Md, 20-21 August 1990 {dates are incorrect on the STScI web page}, eds. White and Allen).  Consider, for example, the case of the "fascinating 'ladder-like' structure in the ejecta of Eta Carina as found in deconvolved images from HST's spherically abberated WF/PC-1 planetary camera as recounted in this STScI Press Announcement.  Once the aberration corrected WF/PC-2 was used to conduct follow-up observations, without the "necessity" of deconvolution that, and other "structures" disappeared, being recognized as image artifacts introduced in the deconvolution process.


The Proof is in the Pudding.
"Proof" often comes practically, so, when two (or more) different methods, applied to two (or more) different data sets of the same observed phenomenon, agree in their results we can (cautiously) declare "victory".  In this case, the above "final" image is directly compared with that of Drückmuller (as it will appear in a paper in the journal of Solar Physics).  As a detail, The Whiddon inner corona image is reproduced at the same image scale and orientation as the Drückmuller image as shown below. Specifically, the Whiddon images (as above)  is reduced in scale, i.e., resampled to fewer pixels (56.2%) than the actual pre and post processed image data, to enable a direct comparison with the Drückmuller image as as it will appear in the paper in preparation.



Inner coronal imagery from the World Explorer (Drückmuller) and MS Paul Gauguin (Whiddon/Schneider)

The coronal feature at the "3 o'clock" position is of particular interest, and is recovered from a single non-composited (!) image with higher image contrast.  The Drückmuller image (left) provide somewhat higher spatial resolution (particularly at the lowest image contrasts, and of course at the smallest sampling scales), and higher S/N at larger radii. In the Drückmuller image the directionally uncorrelated highest spatial frequency components in that image are retained, but are filtered in the Whiddon image.  This can be done effectively in the Drückmuller image thanks to the higher input S/N resulting from the finer effective input-image sampling gained by the 32-image compositing.  None-the-less the amount of (spatial) detail which can be recovered from a single image by the filtering method described here is significant.


It is also QUITE instructive to BLINK THESE TWO IMAGES as well.

The astrometric registration between these two images is imperfect. The rotation and scale were  determined empirically by mapping half a dozen nearly-unresolved prominence features in common between the two images - but those features are time variable.  None-the-less, we MAY be able to see proper motions due to the highest flow velocities in some coronal features... even over the appx. only 3.5 minutes between images!

The Drückmuller image was obtained at appx: 19:51:34 UT
The Whiddon/Schneider was obtained at appx: 19:56:04 UT

Note, for example, note the change in angle of the "opening V" at the bottom of the 3-o'clock loop  (which appears to move slightly outward (to the right) with time compared to the inner part of the top of the arch (which appears to remain stationary; put an arrow cursor on that as it blinks).  It is hard to attribute that to an astrometric error as the differential position between the locations of those two spatially close features is very small.


TSE 2005 ECLIPSE IMAGING from some of the of other umbraphiles on the M/S Paul Gauguin.


Left to right: Charles Cooper (out of uniform), Roland Burley, Robyn Small, Joel Moskowitz, Jay Friedland (the really tall guy) , Jean-Luc Dighaye, Glenn Schneider (in partial eclipse by J-L), two anonymous folks in background, Michael Gill, John Beattie, Craig Small. {click linked names to see their images}


Glenn Schneider, Steward Observatory, The University of Arizona
Last Update: 15 November 2005