Determining pattern–process relationships in heterogeneous landscapes

Landscapes are now being altered at unprecedented rates (Forman and Alexander 1998), resulting in the loss and fragmentation of critical habitats (Gardner et al. 1993), declines in species diversity (Quinn and Harrison 1988, Gu et al. 2002), shifts in disturbance regimes (He et al. 2002, Timoney 2003), and threats to the sustainability of many ecosystems (Grime 1998, Simberloff 1999). Because the ecological consequences of landscape change are difficult to measure, especially at broad spatial and temporal scales, the quantification of landscape pattern has often been used as an indicator of potential biotic effects (e.g., Iverson et al. 1997, Wickham et al. 2000). It is hardly surprising, therefore, that the development of methods to measure landscape pattern has become an important endeavor in landscape ecology (see O’Neill et al. 1999 for a recent review). Numerous landscape metrics have been developed and applied over the last 15 years or so, but relatively few studies have been successful in using metrics to establish pattern–process relationships at landscape scales. The first landscape metrics paper (Krummel et al. 1987) attempted to do this by presenting the hypothesis that the shape of small forest patches should be affected by human activities while large patches should follow natural topographic boundaries. The analytical results showed that this was the case, but causal relationships were never experimentally confirmed. The prospect of this first study stimulated the rapid development of additional indices (see O’Neill et al. 1988, Haines-Young and Chopping 1996), with progress in this arena frequently reviewed (e.g., Gustafson and Parker 1992, Riitters et al. 1996, Hargis et al. 1998, Fauth et al. 2000, He et al. 2000, Tischendorf 2001) and computational methods codified (e.g., Gardner 1999, McGarigal et al. 2002). However,