For over a decade there has been increasing interest in considering the spatial dimension of ecological processes. Researchers and practitioners are interested in spatial landscapes created by diverse phenomena. Attention has been directed towards landscape characterisation, especially on their quantification. In order to characterise a landscape, one is interested in the spatial entities generated by a given -ecological- process. In addition to problems associated with defining common attributes of these entities, the definition of transition zones, or boundaries, between objects is also a complex matter. Because numerous natural phenomena create ill-defined spatial objects it is therefore difficult to determine where a region begins and where it ends.

Quantifying a landscape raises the question of how to define spatial entities if there is uncertainty in the boundary location. The degree of fuzziness of a boundary is intrinsically linked to the processes involved in the formation of a particular spatial entity. Hence, not all boundaries can be analysed the same way because of their different properties. The variation of the width of the transition zones between regions makes the choice of scale of analysis important.

With the increase of computer capacities and the development of powerful spatial analysis techniques, geographical information systems (GIS) are a powerful tool in boundary detection. In remote sensing, filters have been developed to facilitate the interpreter's work and to minimise the uncertainty associated with the interpretation. The choice between different filters, that have been developed for remote sensing and that are now incorporated into GIS, depends on the character of the data and/or the type of analysis. Some of them aim to enhance edges in an image (e.g., Laplacian filters). Filters work by giving to a pixel in the resulting image, a value calculated as the sum of the product of the original values within a given window. In other words, a filter multiplies the cells within a window and the sum of those products is assigned to the centre pixel (same location on a new image).

The result of using a Laplacian filter is that the value of the centre pixel in a new image is increased in relation to the neighbouring pixels. This procedure allows a better recognition of changes in the data set. Furthermore, filters can be used with different window sizes, thus changing the size of the surrounding area considered in the analysis. Modifying the window size is the equivalent of changing the scale of analysis; the processes are being analysed more or less locally.

This experiment addresses the question of boundary recognition using Laplacian filters of different window sizes. The purpose is to take into account the level of autocorrelation within the patches of the landscape and the difference between the averages of two regions, and assess how such parameters can influence the ability of the filter to detect a known boundary.