Solar System
Shoemaker-Levy 9: Rebuilding Hubble's 1994 Jupiter Impact in Color
Black-and-white frames minutes apart, a planet that turns fast, too few blue filters, and a 3D model built to undo Jupiter's rotation.
In July 1994 the fragments of comet Shoemaker-Levy 9 slammed one after another into Jupiter, and Hubble was watching. It was one of the telescope's first great public successes. More than twenty-five years later, a documentary for Hubble's 30th anniversary, the BBC and Discovery Science film Hubble: Thirty Years of Discovery, needed something the original releases never offered: a smooth color sequence of an impact as it happened. Building one took about ninety new images and a technique borrowed from 3D animation.
The impacts of July 1994
The comet had broken into a string of fragments, labeled with letters, and they struck Jupiter's southern hemisphere over about a week. Fragment A hit on July 16, E and F on July 17, H on July 18 and Q on July 20. Each left a dark scar in the clouds. By July 22 the Wide Field and Planetary Camera 2 could record a whole row of them across the disk, as in the picture at the top of this page.
The largest single event was the impact of fragment G. The documentary wanted to follow it frame by frame: the fireball plume rising over the edge of the planet, spreading outward, and collapsing into an expanding wave on the cloud tops. The D/G complex, where fragments D and G struck on July 17 and 18, was later struck again by fragment S on July 21, and the gray panels above show it smearing into Jupiter's winds over the following month.
Why the original frames resist color
Hubble takes one picture at a time through one filter. A color image needs at least three, normally red, green and blue, and during the G impact sequence there were not enough of them to go around. The whole set held only one blue-filter frame. Sometimes an ultraviolet frame could stand in for blue, though it does not look quite the same. In some stretches only red and green were available, and the color of the cloud bands and the Great Red Spot had to be added back by hand to keep them from looking wrong.
The bigger problem was rotation. Jupiter turns once in about ten hours, and the frames were several minutes apart. In that time features move noticeably across the disk, so frames taken through different filters do not line up. To combine them, the planet's rotation has to be undone: each frame must be warped so that the clouds sit where they would have been at one chosen moment. For a single still image that can be done by nudging pixels in an image editor. For roughly ninety frames, delivered within a few months, it was hopeless.
The usual planetary-science answer is a map. Rolling the globe out into a flat cylindrical projection, aligning the maps and wrapping them back onto a sphere removes rotation neatly. But a map assumes a fixed surface, and here the subject was a plume climbing above the limb and an impact scar spreading rapidly. Anything that turned the scene into a static map would erase the very events the film needed.
A model of Jupiter in 3D software
The solution was to treat the planet as a 3D object. In Blender, a free and open-source 3D program, a Hubble frame was projected onto a spheroid with Jupiter's shape, aligned to the image. An orthographic camera, one without perspective, then rendered the scene so that the output matched the original Hubble pixels. Rotating the spheroid by the right amount warped the projected image exactly as Jupiter's rotation would have, ready to match any other frame.
Getting the angle right was arithmetic. A spreadsheet counted the minutes between each pair of observations and multiplied by Jupiter's rotation per minute. The renders were then made one by one with those values. The first version worked within a day; the render came out slightly darker than the original, a side effect of how the software handles color, but otherwise matched it closely.
Two problems remained. A rotated sphere drags its shading and its cut-off edges along with it, and near the limb the planet looks wrong. Rotation matters least at the edge, so the fix was to use the unrotated image for the outer rim and feather it into the warped interior. Alignment was the other issue, since Jupiter was partly cut off in many frames and its full outline had to be guessed. A wire-mesh grid generated with an online Jupiter Viewer tool gave an accurate outline to fit.
Putting the impact back
Because the color layers came from different moments, the plume did not agree between them. Where the color frames showed a rising plume, the frame being colored might already show the flattened wave. So the plume was removed from the warped color version, the planet's shading was corrected, and the wave from the original frame was blended back in, with its brightness adjusted so that it sat naturally on the colored planet.
The wave's color was a judgment call, since no frame recorded it fully in color. It was given a faint yellow tint; Heidi Hammel, one of the principal investigators of the 1994 observations, described that as a reasonable guess. A slight blue edge was added at its top, where the material rose above the atmosphere, and a reddened base, like sunlight low on the horizon. Nobody can know exactly how it looked, and the tints are small enough that a viewer would not notice them unless told.
The finished sequence still has small flaws, left in when further polishing stopped making a visible difference. Methods like this one sit at the edge of what counts as processing: the Hubble data underneath stay real, while filling color gaps and re-placing a plume are interpretive steps. The same derotation problem in later Jupiter and Saturn images is covered in What's Up with Jupiter and Saturn?.
Questions readers ask
When did comet Shoemaker-Levy 9 hit Jupiter?
The fragments struck over about a week in July 1994. Fragment A hit on July 16, E and F on July 17, H on July 18 and Q on July 20. The largest event was the impact of fragment G. Hubble recorded the impact sites with its Wide Field and Planetary Camera 2.
Why could the Hubble frames not simply be combined into color?
Each frame was taken through one filter, minutes apart, while Jupiter rotated about once every ten hours. Features moved between frames, so they did not line up. There were also too few blue-filter frames, and some stretches had only red and green.
How does 3D software help undo Jupiter's rotation?
A Hubble frame is projected onto a Jupiter-shaped spheroid and rendered with a camera that has no perspective, so the output matches the original pixels. Rotating the spheroid by the angle Jupiter turned between two frames warps the image to line up with the other frame.
Is the color of the impact plume real?
Only approximately. No frame recorded the plume and wave fully in color, so they were tinted by judgment: a faint yellow, a slight blue edge at the top and a reddened base. One of the original principal investigators judged the yellow tint a reasonable guess.