Currently, the toolkit's animation techniques are usable on any data layers which are scaled between 0 and 1, e.g. the verification and fuzzy layers in the serialised files.
Single layers can be 'stepped through' in sequence in a serial animation, using a moving threshold to mask out areas on the basis of their value. This picks out patterns and structures in the data, and can also point out correlations between data layers.
Sets of data layers can be combined together in a random
animation. This is particularly useful where the data layers complement one
another (e.g., ground cover proportions, where there are a set number of
possible landcovers, covering varying proportions of each pixel, and within each
pixel, the proportions sum to 1). A random animation allows you to pick up the
different possibilities for each pixel.
Random animation uses a set of fractional memberships to generate a set of alternative likely views, and show them in sequence. For example, using the 'Random animation' option on the display bean, and selecting all the fuzzy membership layers will give an animation looking like this,
which can be started, stopped and altered in speed using the menus at the
For every pixel, the data from all the layers affects that pixel's colour in each frame. This animation shows predicted ground-cover proportions, and the fuzzy memberships in a pixel can be visualised as a stack, adding to 1. For every frame, a random number between 0 and 1 is generated, and the colour of the pixel depends on where in the stack the number falls. A dominant landcover, with high membership, is more likely to be 'chosen'. Running the animation shows areas where certain classes are dominant or more likely. It also picks out areas of characteristic mixing, such as the bottom right of the image, which contains many small features made up of the landcovers 'tile', 'tarmac', 'deciduous' and 'slate'. The aerial photo shows that this is a typical urban area, with small features such as roofs, roads and trees falling below the scale of the pixel.
These animations are produced pixel-by-pixel - there is no spatial
autocorrelation built into the visualisation.
Serial animation steps through a set of continuous values, according to some rule, (e.g., showing increasing membership values as time goes on). There are 3 options by which serial animations can be produced, using the 'Serial animation' option on the DisplayBean, and all methods can be used on one or more layers of data at a time;
In this method, an ordinary grey-scale image is subjected to successive 'alpha cuts' with a progressively higher threshold, so that only the pixels above this threshold value are displayed. The grey scale values still represent actual membership values, so that darker grey areas disappear first. This picks out spatial structure in the image, as mixed or uncertain areas are sequentially masked out.
The slider bar is used to adjust the threshold for all the animations at once, so that they can be compared. The 'ascending' and 'descending' buttons will show all frames in sequence - i.e., with the 'ink' moving up and down over the surface.
Here, there is also a threshold, but only pixels whose memberships fall within a fixed distance of the threshold are shown (e.g., between 0 and 0.05, or between 0.05 and 0.1). In this case the pixels are shown in white, on a black background. Using this option, a variety of data layers can be selected, and only the pixels of a certain value shown in each. Again, the animations are controlled by a central slider bar.
From our area in Leicester, a random animation using all the verification
layers looks something like this.
(Produced using 'Random animation', selecting all verification layers, and using the default 30 frames and 3 frames per second).
The animation of the verification data shows genuine mixing within satellite pixels (see the corresponding aerial photograph on the right), and illustrates the limits on certainty in thematic classifications. This animation also shows characteristic types of mixing which are associated with areas of particular land use. (E.g. the mix of tarmac, tile, and vegetation on the bottom right, which corresponds to an urban residential area).
A similar random animation, produced from all the fuzzy membership layers, shows how much more noisy the fuzzy classification is, and how spectral confusion leads to misclassification. (The white stripes in this image are areas masked out of the classification).
Even if we could clearly locate and identify the pixel locations on the
ground, and identify all the elements within that square, there would still be
bias in the spectral signature of this area, because the satellite actually
samples a circular area of land, and takes more account of the centre of this
circle than of the edges.
Try generating your own random animation from the verification and fuzzy layers in the example dataset. They should look something like this:
NB - the fuzzy membership layers have a default colour of black, so you will
need to edit their colour using 'Customise'.
BACK to menu
NEXT - Graphs