The data employed for this experiment were simulated landscapes. A simulator based on the conditional auto-regressive model (Cressie 1993) was used to create several types of landscapes from the same original image by specifying different values for statistical parameters. The original image consist in a raster of 50 x 50 pixels with two homogeneous regions (Figure 1).

For each region the simulator allows the specification of a given value of spatial autocorrelation (SA) and keeps constant the average of each individual region. Thus, the value of SA is known as well as the difference between the regional averages (as the average of each region can be specified and is keep constant throughout the landscape, it is possible to determine the difference between the two regional averages). For this experiment, three combinations of SA where used over both regions: (0.0, 0.0), (0.9999, 0.9999), (0.0, 0.9999). The difference between the regional averages was equal either to two or ten. Therefore the combination of all these parameters leads to six types of landscapes. For every class of conditions, the simulator was run twenty times, creating twenty landscapes.

As the original image is made of two perfectly homogeneous regions (i.e., equal values), the exact location of the boundary is know. The use of a simulator aims to give to the two homogeneous regions a more natural character by adding variability with know averages and SA.