This is intended to be a list of the "best" data available for rendering the planet Mars. I write from the point of view of a 3D artist wanting to create images that are visually realistic, although perhaps not 100% scientifically accurate (so practices like painting out gaps in the data set in Photoshop are allowed)

(All maps are in spherical projection unless otherwise indicated)

The MOLA dataset is a global height map of Mars. It is only available in tiles, but one can assemble a single whole-planet map fairly easily. Officially the height values represent deviations from a high-order spherical harmonics approximation of Mars, but you can cheat by using a simple ellipsoid, and still get good results.

Comments: The resolution is spectacular; tiny craters and valleys are visible down almost to single-pixel scale. The global height map is free of artifacts and noise at large scales, but at smaller scales (features a few pixel across) there are little artifacts that look like parallel scratches along the surface. I think these are the result of uneven spacing in orbital passes over the terrain. Also, terrain at very small scales has a "blocky" appearance due to the rectangular sampling pattern. These artifacts can be mitigated by manually retouching the height map and/or adding higher-frequency texture.

The highest-resolution, whole-planet color map of Mars is a mosaic from the Viking mission. It's very hard to find this version on the web; most sites only have downsampled versions.

Comments: The Viking mosaic is full of artifacts, unfortunately. There are significant lighting discontinuities across different parts of the image, with eye-catching sharp boundaries between them. There is a lot of "double vision" where super-imposed mosaic tiles don't quite line up. There are significant discrepancies in feature locations between the Viking mosaic and more recent data sets like MOLA altimetry and MOC images. Our internal version of this image has been retouched extensively to smooth out the discontinuities and adjust feature positions to match MOLA topography.

The USGS has produced a newer global image called "MDIM2" which matches MOLA much better, except it's monochrome and there are still visible seams.

NEW: Kees Veenenbos has informed me of Mario Rossi's Space Graphics site, which advertises high-resolution (46K) color Mars maps, both with and without topographic shading. These maps are supposed to line up accurately with the MOLA height map. I wrote Mario about obtaining them; it's not clear how to download or purchase full-resolution maps from his site.

If you are willing to forego color, a much better whole-planet image is available in the global MOC mosaic from Malin Space Science Systems. The globe is split into 30 tiles of varying resolution, including special polar tiles for the north and south poles. A fully-assembled mosaic would be 92,160 x 46,080 pixels.

Comments: I have only limited experience with this data set, but it looks mostly free of artifacts. The images are intentionally low contrast, so you will probably have to stretch the dynamic range. It looks like lots of dust was in the atmosphere when some of the images were captured; features such as Olympus Mons are somewhat obscured. Malin claims that the MOC mosaic lines up very accurately with MOLA topography, but I haven't verified this.

The next step up in resolution (and down in regional coverage) from the MOC global mosaic is the THEMIS visible and IR dataset. (You can usually get away with using daytime IR images as if they were visible, since you're mostly looking at topographic shading, not surface albedo).

THEMIS images come in long, thin strips. The IR images are about 100m per pixel and the visible images are about 20m per pixel - higher resolution but with a narrower field of view. There are IR images for almost the entire planet's surface. Visible coverage is sparser, although most of the famous Martian landmarks are covered well. Usually we start our texture maps by laying down low-resolution IR images and then filling in high-resolution visible images where possible.

Comments: Image quality ranges from excellent to totally unusable, with most images falling somewhere in between. THEMIS images usually show some amount of "step-ladder" artifacts running across the long strip, which are difficult to paint out. White and black levels are not consistent across images, so you will have to match them up to make a seamless mosaic. Fortunately the images are not gamma-encoded so this can be accomplished with a simple levels adjustment. The visible images are map-projected so you should not have trouble lining them up with MOLA topography. The IR images are NOT map-projected; you will have to skew them into place.

These images are similar to the visible THEMIS images, only higher resolution (<10m per pixel) and with a correspondingly smaller footprint. Coverage of Mars is pretty sparse. I am more likely to use MOC images to build generic texture maps rather than model a specific area of Mars.

Comments: Image quality tends to be pretty good (ignoring the easy-to-remove data drop-outs). Watch out for noise and scan-line artifacts on the enhanced "super-resolution" images.

In most images of Mars, topographic shading, not variation in surface albedo, is the dominant visual effect. This is both a blessing and a curse. A blessing because applying one of these images as a texture map (without any lighting) gives the impression of much more geometric detail than is actually present. A curse because the lighting is "baked" into the texture map -- it's virtually impossible to re-light terrain with a different sun angle. Applying a standard diffuse reflectance to the surface betrays the lack of underlying geometric detail, giving very poor renderings.

An additional problem is that images for neighboring regions may have totally different lighting. This was a huge problem for the Olympus Mons shot in Roving Mars, where the wide-angle MOC image worked great, but all the THEMIS images we wanted to use for close-up detail were lit from the opposite direction, rendering them virtually useless.

Perhaps the solution is to use a shape-from-shading technique to extract a height map from the images, and apply it as a bump or displacement texture?

This really calls for a rendering technique like Terragen's that simulates "nano-scale" displacement. RenderMan-style one-pixel "micropolygon" tessellation is insufficient because it just computes an "average" lighting result for the whole pixel. That's OK for most objects, but not planetary surfaces. Tessellation significantly finer than the area of one pixel is vital to capture how light reacts to the "true" surface geometry.