Ranking habitat patches by contribution to network connectivity: tradeoffs between processing time and spatial realisation

Habitat loss and fragmentation pose a major threat to flora and fauna worldwide. For those species that can move between habitat patches, it is important to consider the extent to which the patches are connected. This can be complex as many species may inhabit a given landscape, and the scale at which each one interacts with the landscape (and consequently its relative ability to move through it between patches) may differ. Nonetheless, designing effective conservation strategies requires prioritising patches for protection. One basis for doing so is the relative importance of patches to maintaining connectivity. This can be estimated by measuring the connectivity of the complete network, and then assessing the extent to which that connectivity changes when each patch in the network is removed in turn. Doing this manually is clearly not feasible for a landscape comprised of more than a few patches. Obviously modeling landscapes in this way is inherently spatial. However, the volume of processing required to drop each patch from the network and measure how connectivity changes even when the process is automated creates a trade-off between the extent to which the spatial representation of the landscape is fully realized (spatial realisation) and processing time. For example, graph theory offers a very efficient method to assess connectivity by representing the landscape as a series of nodes (patches) and edges (links between patches) in Cartesian space. This can be extended (‘spatial graphs’) by geo-referencing the nodes. However, unless there is no resistance to how a given species moves between patches, it is necessary to weight the edges between nodes based on some model of dispersal. This can be made more spatially explicit by weighting edges based on least cost paths (distance between patches taking into account the difficulty of the species to move between them using detailed spatial representations of the landscape). However, doing so greatly increases processing times, and may not be feasible to run for large and complex study areas. Further, because of the variation in spatial realization between connectivity models, the results of these models, especially when considering the importance of individual patches to overall network connectivity, can differ substantially. Thus, there is a need not only to be able to automate the ranking of patches based on their relative importance within the network (for which very few automated tools exist), but also to test the sensitivity of the results to the model (and the associated level of spatial realization) that is used. If patch rankings do vary considerably between models, it would make sense to use a model comparison approach to generate the final patch rankings (as is frequently done in other disciplines where considerable model uncertainty exists such as fire spread modeling or global climate change modeling) before making conservation decisions. This paper presents the Habitat Connectivity Research Software (HABCORES), a software toolbox written for ArcGIS 9.2, which incorporates three different modeling approaches to habitat connectivity that lie along the continuum of possible spatial realism. Preliminary testing of the tool for a case study of koala habitat in south-eastern NSW indicates that different model approaches yield quite different patch ranking results. Future work will include assessing the sensitivity of model results to key parameter settings, incorporating additional connectivity models into the HABCORES framework, conducting further tests of patch ranking sensitivity to the model choice using neutral models and testing connectivity predictions from the model for one or more case study species against field data of individual animal movements in a test landscape.