Placemark Clustering: GeoCubes and Barry Hunter

I've discussed clustering placemarks before but thought a revisit was in order since coming across Barry Hunter's work which is partly based on GeoCubes.


Since Barry develops the idea (with a neat little web form in a placemark which I like) from GeoCubes (design 1) I'll just comment on this GeoCubes visualization.

The visualization shows placemarks (it doesn't matter for this discussion what they represent) collated into square cells. The number of placemarks is visualized by a colored cube and a number written on the cube.

Pros:

1. Clustering Need: You can't render thousands of placemarks on screen at any one time so they must be clustered in an easy to understand way. GeoCubes does this.
2. Grid: In the cubes visualization shown placemarks are collated into a grid which is easily understood (other systems collate placemarks into areas where the boundaries are not known - much less comprehensible by the user)
   
3. Placemarks and Visualization: When you fly into the screen the cubes dissolve to show the placemarks themselves. Thus you can find the exact location of a single placemark.
4. It does all this automatically
   

Cons:

1. Figures on screen distracting: The mind processes numbers serially but color in parallel so picking out all the numbers between 500-800 in the screenshot is slow but you can almost instantly identify the 7 orange squares. This is because the mind processes colours in parallel (try it for yourself). So it would be better to have the number text revealed by a mouse rollover or click but otherwise be invisible.
   
2. Seamless Cubes: The size of the cubes shows the density of placemarks as well as the color of the cubes, it would be more useful if the cubes went all the way up to the boundary of the cell as then cells with the same density would merge into one seamless, color blob. Barry tells me the geocube guys are thinking about this already.
   
3. More Intervals: The number of the placemarks per cell is split into 3 intervals ( 1-100, 101-1000, 1000+). It would be better to have a color blend with more intervals as more information would be immediately apparent (see parallel processing of colour in [1]). Good blends of colors can be gained by colorbrewer which I review here.
   
4. Density vs Count: As you zoom in the cells split themselves into smaller cells so that the placemark count goes down. This causes a change of color in the cells but the density of the placemarks per square km hasn't changed. I think it would be better if the cell colour illustrated placemark density as the color would remain more constant. To see what I mean compare select design #1 at geocubes and zoom in. Compare that to zooming in on this pizza resturant map (only imagine at the lowest altitude the heat map dissolves and leaves just placemarks).
   

If you've been watching carefully you will see that if all the changes I suggest were put in place we would end up with pretty much 'My Solution' outlined in my original post.

Barry's response to the review was positive and he's thinking of producing an experiement in line with my suggestions. I hope that it happens!