Validation of functional connectivity modeling: The Achilles’ heel of landscape connectivity mapping

Abstract Modeling functional connectivity in altered landscapes is one of the growing fields of expertise in landscape ecology, and many research teams have proposed different methods to evaluate it for a wide range of species. However, very few have empirically validated the efficiency of such models in discriminating real corridors from theoretical ones. Models that are not validated or those only based on structural connectivity could result in inefficient management decisions. Moreover, validation could potentially reveal that functional connectivity differs between focal species and spatial scales. Here we empirically compared different validation methods for two commonly used connectivity models applied to two cervid species (i.e. moose Alces americanus and white-tailed deer Odocoileus virginianus) during a road enlargement project. For both species, we built functional connectivity maps using CircuitScape (circuit-based model) and LinkageMapper (least-cost path model). We then validated them empirically using four different metrics: density of cervid-vehicle collisions, distance to the nearest wintering ground and detection rate calculated with automated cameras and with sand traps. Validation was carried out at various spatial scales (150, 500, 1000, 1500, 2000 and 2500 m). The circuit-based models performed better at identifying functional corridors of connectivity for moose. Validation strength differed greatly between the four metrics used, and the spatial scale at which the correlation between connectivity and data was assessed had little effect. Our study emphasizes the importance of validating functional connectivity models to provide the best decision-making tools.