I invite anyone with interest in this image series to check out J Maki et al. for their explaination of the autoexposure algorithms:
http://robotics.jpl.nasa.gov/people/...03JE002077.pdf

This is a great example of the delicate balance they're trying to obtain when picking how to frame an image...this image was, i'm betting, misframed. In short:

When they use the autoexposure function, the image waits until a critical % (pixel fraction) of ccd bins reach a specific pixel value (DN threshold) or higher. The problem in the above image, as opposed to other similar images, is the very low % of the image that has sky in it. If the pixel fraction were 25%, and the sky covers less than 10% of the image, then the 10% sky would max exposure long before the pixels of ground reach the autoexposure DN threshold. This leaves the image exposing long after the ccd bins for the sky portion have filled, and begin to bleed over in the nearby bins. Most other images of the horizon line have been taken at least ~20 % of the image given to the sky, so that the sky brightness is the trigger of the end of the exposure.

You can see in the earlier, L2 L3 L4 frames, there are bright (in that filter's color) features in the foreground. They would contribute to the pixel fraction, and help cut off the exposure before it goes too far. The later L5 L6 L7 frames (green-blue) do not have bright features in the foreground, so the entire pixel fraction has to come from the bright horizon (and those bins which are bled into).


Its also good evidence for the nature of the light coming off the ground and sky. The filters which have the most bleed represent those which are in the blue end of the spectrum, lending more credence to Mars' consistant redness. There isn't enough blue in the scene to cut off the exposure until long after the inherient brightness of the sky overwhelms the nearby bins.