Effects of uncertain cost-surface specification on landscape connectivity measures

Estimates of landscape connectivity are routinely used to inform decision-making by conservation biologists. Most estimates of connectivity rely on cost-surfaces: raster representations of landscapes in which cost values represent the difficulty involved with traversing an area. However, there is considerable uncertainty in the generation of cost-surfaces that have not been widely explored. We investigated the effects of four potential sources of uncertainty in the creation of cost-surfaces: 1) number of landscape classes represented; 2) spatial resolution (grain size); 3) misclassification of edges between landscape classes; and 4) cost values selected for each landscape class. Following a factorial design we simulated multiple cost-surface pairs, each comprising one true surface with no errors and one surface with uncertainty comprised of some combination of the four error sources. We evaluated the relative importance of each source of uncertainty in determining the difference between the least-cost paths (LCPS) costs and resistance distances generated for the true and erroneous cost-surfaces, using four model evaluation metrics. Errors in the underlying geospatial layers produced larger inaccuracies in connectivity estimates than those produced by cost-value errors. Incorrect grain size had the largest overall effect on the accuracy of connectivity estimates. Though the removal of an element class was found to have a large effect on the configuration of connectivity estimates, and the addition of an element class had a large effect on estimates configuration. Our results highlight the importance of minimising and quantifying the uncertainty inherent in the geospatial data used to develop cost-surfaces. (C) 2016 Elsevier B.V. All rights reserved.