Here's a video which summarises some data visualisation I've been doing recently. The rest of this post explains the story behind it.

A common analysis tool for spatial data is "patch statistics". The idea behind this is to divide a map up into connected clusters of the same type - called patches - and then compute statistics for these. This is quite similar to the blob detection used in vision based multitouch projects.

As part of the work I'm doing, we wanted to look at how likely certain areas are to be developed. We wanted to separate out areas which would definitely be developed, definitely not be developed and those which were "up for grabs".

There are several different scenarios about future development, which each provide a coherent set of parameters for the model. The model is also stochastic, so there are 30 runs for each scenario to give probabilistic futures. The task is to take the results from many runs under each condition, and produce a map showing probabilities of development.