Linking Models of Species Occurrence and Landscape Reconstruction

Ecologists and land managers around the world are charged with first arresting and then reversing declines in native species. Revegetation has been proposed as one of the mechanisms by which landscapes can be rehabilitated to support viable populations of native wildlife. Because large-scale revegetation often proves to be technically difficult and costly, it is critical to evaluate the likely outcome of alternative proposals for landscape reconstruction. Here we describe a new approach for examining the potential effects of spatially extensive ecological restoration on species of concern. Our method links validated models of species occurrence with GIS-based models of various revegetation scenarios to estimate the range of biodiversity responses under each option. Explaining and predicting species occurrence long has been a major goal in ecology, conservation biology, and wildlife management (Rosenzweig 1995, Mac Nally 1995, Bell 2001). There are many possible ways to predict species occurrence. Traditional ‘habitat modeling’-predicting occurrence as a function of resource requirements, such as food sources or nesting sites-may have a high probability of success (Hanski 1999, Miller and Cale 2000), but obtaining such data can be expensive, particularly over extensive areas. Therefore, predicting species occurrence as a function of environmental variables that can be quantified easily, at small spatial grains, and over large areas, is appealing (Austin et al. 1990, Guisan and Zimmerman 2000, Jackson et al. 2000). We recently developed a statistically rigorous framework for examining the generality of predictors of species occurrence using an iterative process of model building, testing, and refinement (Fleishman et al. 2001, in press). We make extensive use of Bayes-based methods, which facilitate more detailed and practical assessment and improvement of predictions than conventional approaches (Ellison 1996, Sit and Taylor 1998). Our framework seeks to identify predictors of species occurrence at grain sizes on the order of several km2 over extents of 100s to 1000s of km2. This corresponds to the scale at which many landuse decisions must be made. To be useful, the predictions of species-occurrence models must be tested using explicit standards (Guisan and Zimmerman 2000, Jackson et al. 2000). We test our models-which effectively are hypotheses about predictors of species distributions-using independent data that were not used to build the models (Fleishman et al. in press). The process of generating and testing model predictions increases our understanding of relationships between organisms and environmental variables and contributes to the scientific foundation for regional conservation and management (Mac Nally and Bennett 1997, Hawkins et al. 2000, Mac Nally et al. 2000). Species-specific occurrence modeling has been employed widely in the past (Braithwaite et al. 1989, Lindenmayer et al. 1990, Scott et al. 2002), but occurrence models rarely have been linked with GIS-based models of alternative management strategies or revegetated landscapes (Bennett 1999, Marzluff et al. 2002). The alternatives we develop are based on ecological vegetation classes, which are defined as one or more similar floristic communities that exist under a common regime of ecological processes and that are linked to broad landscape features (Muir et al. 1995). Because ecological vegetation classes are closely connected with broad-scale topographic, edaphic, and climatic variables, they are a useful link between vegetation and landscape-scale planning and management. Alternatives vary with respect to the amount and spatial configuration of different ecological vegetation classes. Our approach recognizes that not only is there a spectrum of habitat quality (McIntyre and Barrett 1992), but also that animals respond to more than one vegetational community (Mac Nally et al. 2002). Connecting occurrence models with revegetation models allows us to estimate the quantity and distribution of suitable habitat for each species that would be available under each scenario. By treating models for individual species as probabilistic, we can generate ranges of outcomes (i.e., confidence intervals) for each alternative. Thus, we can gauge the overall potential of each alternative to achieve specified ecological objectives.