Can landscape indices predict ecological processes consistently?

The ecological interpretation of landscape patterns is one of the major objectives in landscape ecology. Both landscape patterns and ecological processes need to be quantified before statistical relationships between these variables can be examined. Landscape indices provide quantitative information about landscape pattern. Response variables or process rates quantify the outcome of ecological processes (e.g., dispersal success for landscape connectivity or Morisita's index for the spatial distribution of individuals). While the principal potential of this approach has been demonstrated in several studies, the robustness of the statistical relationships against variations in landscape structure or against variations of the ecological process itself has never been explicitly investigated. This paper investigates the consistency of correlations between a set of landscape indices (calculated with Fragstats) and three response variables from a simulated dispersal process across heterogeneous landscapes (cell immigration, dispersal success and search time) against variation in three experimental treatments (control variables): habitat amount, habitat fragmentation and dispersal behavior. I found strong correlations between some landscape indices and all three response variables. However, 68% of the statistical relationships were highly inconsistent and sometimes ambiguous for different landscape structures and for differences in dispersal behavior. Correlations between one landscape index and one response variable could range from highly positive to highly negative when derived from different spatial patterns. I furthermore compared correlation coefficients obtained from artificially generated (neutral) landscape models with those obtained from Landsat TM images. Both landscape representations produced equally strong and weak statistical relationships between landscape indices and response variables. This result supports the use of neutral landscape models in theoretical analyses of pattern-process relationships.