| Bad Example |
| Good Example |
Bad Example
Description
This image is from http://www.matemagic.it/image_gallery.htm. In particular it is described as a "Triangular mesh reconstruction from generic plane dataset" that has been produced by GPR (Graphic Post-processing Resources), a tool for 2D and 3D visualization
Flaws
There are a number of areas that have room for improvement in this visualization. First, there is no labeling whatsoever, either positional (such as axes) or of the scalar color. Thus, the viewer gains no information as to numerically what are where they are looking at on the dataset.
The next main issue is the use of depth in the image, or rather lack there of. It's unclear whether you are viewing a planar surface or if there is a 3d dimension of depth from the image. One improvement would be to include some form of shadowing on the surface to indicate higher/lower values. Also, it is nearly impossible to discriminate values on the back of the image because they are greatly distorted, and partially occluded by the mesh surface.
Finally, I find the choice of colors to be non-intuitive. They range from cyan to red and then to purple. However, one could argue that the purple should be on the lower end of the range, or that perhaps you should range from a darker blue up to a red. Plus there are only discrete color changes, although this may be by design or a part of the dataset.
Strengths
I do, however, find the fact that they used some form of color important in this example (even if I find it poorly chosen). Without the color there would be almost no representation of the numeric values.
Also, showing the mesh surface gives some structure to the image. It does help show where the data dips and rises, in a sense. If connectivity, or observing the aspect ratio of some triangulation of the values is important than this image excels in that. Blocking the graph nodes as small, colored cubes, was also a good choice, esp. if knowing the value of a particular node is important.
Good Example
Description
This image is from http://www.stanford.edu/~hpitsch/LESExample.html. It shows the depiction of a Large-Eddy simulation of a turbulent diffusion flame. Also important here is that the colored surface is the actual temperature surface, but there is a black line which is the contour of the stoichiometric mixture fraction (a representative mean/combination of the data).
Flaws
I feel as though there are few flaws in this dataset. However, one potential area way it may be improved is close to the bottom center, where it is hard to discriminate the different values. Perhaps a closer view of that particular region (if it is of interest) may be in order (in fact, on the page, they do show such an image below).
Also, perhaps widening the chart would improve it in many aspects. Because there is such a large gradient of color values, the narrow graph blends them together very rapidly. In particular, it is difficult to identify a temperature value exactly for a particular color, although they have done a good job of providing a scale on the righthand side.
Strengths
There are many merits to this sort of visualization. First, color is used tastefully and provided with a verbose scale on the right to help identify its meaning. Moreover, it is being used in the general red as "hot" and blue as "cold" sense, with higher temperature values being drawn as closer to red, and lower closer to blue. This form is very intuitive, especially considering that dataset is temperature, so they chose a domain specific coloring.
Of course, labeling is done very well, with scales for both the X and Y axes, as well as tickmarks and labels at incrementally increasing values. In addition the color scale is labelled all throughout the range of colors, instead of just on either end, so it does give one a good indication of what each color represents.
Visually, I particularly like two additional aspects. First, the drawing of the contour line of the stoichiometric mixture gives a very good "skeleton" representation of the dataset. Second, they used blending of the colors to show a better resolution of color distribution, although particularly on the outskirts of the dataset, this may just be an artifact of the sampling.