Scaling properties of multivariate landscape structure

Analyses of landscape context is essential for understanding how ecological patterns and processes relate to space. This requires that we quantify variation patterns of different landscape parameters, which may change relative to one another at different spatial scales. Here, we analyzed how statistical relationships of land-use composition parameters changed as a function of extent in 20 real agricultural landscapes. Furthermore, we tested the generality of these scaling relations in numerical simulations using 300 artificial landscapes. We analyzed proportions of artificial habitat types at different extent and compared these patterns with three dominant habitat types in real landscapes (forest, arable land and grassland) at four spatial scales (quadrates of 1-4 km). Both real and simulated landscapes showed that variance of landscape parameters (data differentiation) decreased and their correlations (data consistency) increased as scale increased, thereby suggesting general scaling laws. The potential statistical impact of these scaling relationships is revealed in simultaneous analyses of variation of (local) site parameters of 20 arable fields and their surrounding landscape context. At small and medium extent (quadrates of 1-3 km), variability of local site parameters (e.g. fertilization, pH-value) was high relative to those of landscape parameters. In contrast, at large extent (quadrates of 4 km) variability of landscape parameters was greater than that of site parameters indicating a fundamental shift in the relationship between these sets of parameters at different scales. Hence, it is clear that there is a high risk of artefactual correlations in hierarchical multi-scale landscape analyses when ecological data are related to the landscape context. Accordingly, there is a necessity for multi-scale analyses in landscape ecology to accurately evaluate the relative importance of landscape context at different spatial scales. (c) 2005 Elsevier Ltd. All rights reserved.