BEHAVIOR OF A MULTIVARIATE ALGORITHM FOR ECOLOGICAL EDGE-DETECTION

Ecologically important edges are best characterized by multivariate data, and represent organizational changes of varying intensities. Squared Euclidean Distances (sed) calculated by a moving window algorithm have paradoxical properties of representing experiences and realities on well studied transects, while at the same time presenting several mathematically undesirable properties. A simulation study was conducted to examine the behavior of the sed metric on data with known properties. Data with known properties were generated from simulation models to represent simple edges and increasing levels of edge and patch complexity. Different organizations of patches show different edge signatures. Some edges in complex patchiness patterns are not detected. Edges become more distinct from background noise as window-width is increased. Edge recognition becomes difficult as edges become smaller relative to background heterogeneity. Simulations provide a way to relate described properties of the sed algorithm to interpretations of field data. © 1990.