This project was funded to understand how, where, and why outputs from landscape connectivity models vary, and to suggest approaches to increase comparability and interoperability of models across Landscape Conservation Cooperative boundaries. We began by compiling metadata from 73 landscape connectivity modeling projects into an online, editable spreadsheet. Using spatial data from a subset of studies included in the database, we conducted an uncertainty analysis to understand how much spatial variation there was among predictions from different landscape connectivity models. Raw outputs from the original models showed relatively little overlap, averaging about 3% across all pairs of studies. However, when a common set of nodes was connected using the original resistance layers from all studies, overlap averaged between 24—35%. A subsequent sensitivity analysis revealed that the vast majority of variation among landscape connectivity models is associated with differences among species, with differences among modeling algorithms and land cover datasets contributing little variation in predictions from connectivity models. Because our sensitivity analysis indicated little variation in landscape connectivity predictions between models generated for the same species using different land cover datasets, we propose that one way to increase the interoperability of independent connectivity models may be to cross-walk resistance scores to a standard land cover product. We suggest that the National Land Cover Dataset is the ideal product for integration of landscape connectivity modeling projects. The four key recommendations for increasing comparability and interoperability of landscape connectivity models emerging from our work are: (1) coordinated modeling of multiple species across LCC boundaries; (2) identification of common nodes across studies or portions of the landscape; (3) use of the national land cover dataset to score resistance across the landscape; and (4) use of least-cost path models to connect nodes. We demonstrate our proposed approach with 27 models from the connectivity database we compiled. We found that an average of about 50% of the overall original network could be recreated using our standardized approach. From a link-by-link perspective, the standardized approach produces networks that overlap an average of 73% of the original links in each study. The extent to which the standardized approach we propose produces results similar to the original studies is determined in large part by body size of the modeled species, with greater overlap between networks for small-bodied species than large-bodied species. Finally, we used the proposed standardized approach to integrate results from multiple studies of the American black bear throughout the USA. The overlap between connectivity networks in the original studies and standardized network was 32%, a lower-than-expected result that may indicate the difficulty of generating consistent landscape connectivity models for large-bodied species in complex landscapes with large inter-node distance, as was the case for much of our model domain. We suggest that as more data become available to facilitate additional comparisons, it may be possible to better ascertain the combinations of landscape context and species traits where standardized landscape connectivity modeling may be best implemented. Overarching points emerging from our work include: 1) Most landscape connectivity modeling studies take a species approach rather than describing connectivity for habitats or other landscape units, 2) Mammals were the most frequently modeled group of species, 3) The vast majority of studies (75%) use least-cost paths to model landscape connectivity, with circuit-theory based models used in about a quarter of studies, 4) Raw outputs from independent landscape connectivity models show little overlap, 5) When landscape connectivity networks for different species are constructed for a common set of nodes, overlap among networks increases substantially, 6) Least-cost paths provide an intuitive, widely-used framework for developing and comparing landscape connectivity predictions, 7) Species identity contributes most to variation in predictions from landscape connectivity models, whereas the identity of the land cover dataset used to score resistance contributes least to model predictions, 8) Because the identity of the land cover dataset used for modeling contributes little to variation in model predictions, it should be possible to crosswalk resistance scores from individual studies to a common framework using the NLCD, 9) Landscape Conservation Cooperatives can oversee the development of national connectivity maps highlighting connections across LCC boundaries that could be refined at local scales, 10) Applying resistance scores for individual land cover units to a standardized national land cover map produced networks that overlapped an average of 50% of the original least-cost path network from individual studies, 11) On a connection-by-connection basis, the same standardized approach to scoring resistance overlapped an average of 73% of each connection identified in original studies, 12) Overlap between standardized least-cost path networks and original networks is greatest for small-bodied species, and decreases with increasing body size of the species being modeled, 13) Implementation of the standardized approach to create integrated landscape connectivity maps for the American black bear based on data from multiple independent studies resulted in a least-cost path network that overlapped 32% of the area of the original networks, 14) Lower than expected performance of standardized approach, even after considering effects of organism body size on comparability of landscape connectivity results, may reflect the highly variegated landscape and great distance among nodes in the southeastern USA, and 15) Additional tests with more species as data become available may better clarify the potential for the standardized approach we describe to facilitate integration of landscape connectivity modeling efforts across the LCC network.
