This dataset was derived from the California
GAP land cover for this area. It has been reclassified into 4 cover
classes:
Urban (in yellow), Agriculture
(in light green), Scrub (in dark green), Wetland (in blue).
13,784 cells (each approximately 90 meters on a side) are classified as
one of these four classes.
This dataset was perturbed by adding "noise" to each class in the following manner. Each cell has an 80% chance of retaining its value, and a 20% chance of randomly changing to another class. The probability of being transformed into another particular class is weighted so that the resulting noisy map retains the same proportion of each land cover class as the original data. The result is a map which retains the general pattern of the original but with "salt and pepper" inclusions.
The noisy data can then itself be perturbed, introducing more salt and pepper. This process can be repeated until the spatial structure of the original data is lost. This occurs after only a few iterations! The following small images are iterations 1, 2, 5, and 8.
Below is an animated gif which includes iterations from 1-8. Each land cover class approximately retains its original proportions throughout.
Spatial structure is a critical geographical characteristic. Modeling it appropriately is vital for understanding the spatial distribution of environmental phenomena.