Cloud altitudes from stereo (experimental)

Below is an image processed from JunoCam image PJ47_90. It includes highly approximate cloud altitudes in km relative to the 1 bar level (71492 x 66854 km spheroid). This is experimental. The cloud altitude values are computed from a stereo pair using the PJ47_90 and PJ47_91 red filter images. These images were obtained 2 minutes apart.

North is to the upper left. A smaller image with a latitude/longitude grid is also included to show the viewing geometry.

The altitude values are highly approximate and possibly the only fairly safe conclusion from the image is that bright clouds are higher in the atmosphere than dark clouds (this was of course already known). The computed altitude values are highly sensitive to the exact coordinates of the measured points in pixels in the images. Varying x or y in one of the stereo pair images by ± 0.5 to 1 can change the computed altitude value by as much as ~10 km but usually the change is significantly smaller. This means that only a small cloud movement in the 2 minutes beween the images can mess up the results. 1 pixel corresponds to ~7 km horizontally in the highest resolution areas. The corresponding '1 pixel wind speed' is ~60 m/s. In addition to movement, the clouds might also change shape, grow or shrink. A particularly problematic area can be seen slightly above center (points with the values -27, -19, -25 and 0). I also measured several 'pairs' of points where one point is bright and one point is dark. If the cloud movement is similar at both locations and the clouds do not change much the relative altitude difference should be fairly accurate. But needless to say, the bright and dark clouds could be moving at different speeds. Despite all of this, the dark clouds are always at a lower altitude than nearby brighter clouds, implying that the computed altitude values probably make at least some sense.

Interestingly, the altitude values are probably lower in the upper left part of the image. The top of the main clouds is known to be lower in the atmosphere at high latitudes than closer to the equator. These altitude values are consistent with that. However, small systematic errors in the viewing geometry for one or both of the images cannot be ruled out. These errors could result in a slight downward 'tilt' towards upper left. It is also suspicious that the altitude values indicate that most of the clouds are below the 1 bar level; this is very unlikely to be correct.

I have also experimented with generating DEMs from JunoCam stereo pairs but this has not worked well. All of the DEMs I have generated so far have been extremely noisy and usually do not correspond very clearly to what's visible in the images. So I decided to compute the altitude at a few individal points instead. In that case I can manually check all of the points and also (as mentioned above) try changing their pixel coordinates slightly in one image to see how it affects the computed altitude. More caveats - the points are not randomly selected, resulting in possible bias. Also I somewhat arbitrarily rejected a few points where the error estimate was high or the result was obviously wrong for other reasons.