Author's personal copy Parsimony in landscape metrics: Strength, universality, and consistency

Ecologists can be overwhelmed by the number of metrics available to quantify landscape structure. Clarification of interrelationships and redundancy is needed to guide metric selection and interpretation for the purpose of landscape monitoring. In this study we identified independent components of classand landscape-level structure in multiple landscapes in each of three large and geographically disjunct study areas. We used FRAGSTATS and principal components analysis (PCA) to identify independent components of landscape structure, and cluster analysis to group the components. We then calculated the universality, strength, and consistency of the identified landscape structure components. At the class-level we identified 24 independent configuration components. Seven of these components were nearly universal and consistent in interpreted meaning. At the landscape-level there were 17 independent structure components. Eight of these components were universal and consistent. These results indicate that there are consistent combinations of metrics that universally describe the major attributes of landscape structure at the classand landscape-levels. Published by Elsevier Ltd. * Corresponding author. Present address: USDA Forest Service, Rocky Mountain Research Station, 800 East Beckwith, PO Box 8089, Missoula, MT 59807, United States. Tel.: +1 406 546 6379. E-mail address: [email protected] (S.A. Cushman). avai lab le at www.sc iencedi rec t .com journal homepage: www.e lsev ier .com/ locate /eco l ind 1470-160X/$ – see front matter. Published by Elsevier Ltd. doi:10.1016/j.ecolind.2007.12.002 Author's personal copy information (e.g., mean patch size and patch density). In other cases, metrics may be empirically redundant, not because they fundamentally measure the same aspect of structure, but because for the landscapes under investigation, different aspects of landscape structure are correlated. Previous studies have attempted to determine if the major components of landscape structure can be represented by a parsimonious suite of independent metrics (e.g., McGarigal and McComb, 1995; Riitters et al., 1995; Cain et al., 1997; Scanes and Bunce, 1997; Tinker et al., 1998; Griffith et al., 2000; Lausch and Herzog, 2002; Cifaldi et al., 2004; Linke and Franklin, 2006; Schindler et al., 2008). Each of these studies suggested that patterns can be characterized by relatively few components; however, the identified components differed among the studies. This lack of concordance in important landscape structure components raises the possibility that there are no fundamentally important aspects of landscape structure and instead that structure patterns are peculiar to specific landscapes. The apparent lack of fundamental components is, however, more likely a consequence of the fact that the different studies did not use the same pool of metrics and used different methods to identify components. McGarigal and McComb (1995), for example, identified a set of components based on a small number of class-level metrics in 30 landscapes within a single geographical area. Prior to factor analysis, the percent of landscape in the focal class was partialled out to remove effects of class area on landscape configuration. In contrast, Riitters et al. (1995) quantified redundancy of landscape-level metrics from 85 landscapes sampled across a vast geographical space. However, their analysis did not explicitly assess consistency of component meaning or universality of component presence among regions, nor did it separate effects of area and configuration. The overall goal of this study was to identify independent components of landscape structure in three geographically isolated regions and to determine whether the identified components were idiosyncratic or whether components were common across class types and regions. Structural components analyzed in this study are principal components (i.e., composite variables). They are the major independent dimensions of landscape structure that exist among the measured landscapes. We evaluated the importance of each component using three measures: universality, strength, and consistency. Universality is the percentage of classes or regions in which a component is found. Strength is assessed as the average variance explained by a structure component across classes and regions. Consistency measures the stability of component interpretation among classes and regions. We combine the strengths of the previous studies by sampling a large number of landscapes from a variety of physiographic provinces (e.g., Riitters et al., 1995; Cain et al., 1997) and by separating effects of landscape composition from configuration at the class-level (McGarigal and McComb, 1995). We improve on those studies by including a larger number of metrics, by examining landscapeand class-level metrics separately, and by evaluating the importance of landscape structure gradients across classes and regions. Our analysis did not focus on functional metrics that are explicitly related to ecological processes. By definition, functional metrics change in definition and interpretation with changes in the goals and methods of study. We focused on ‘objective’ structure metrics because they are constant among studies. Analyses similar to this, but using functional metrics, would be useful within the context of particular ecological analyses.