Validating network connectivity with observed movement in experimental landscapes undergoing habitat destruction

Maintaining the ability of organisms to move between suitable patches of habitat despite ongoing habitat loss is essential to conserving biodiversity. Quantifying connectivity has therefore become a central focus of conservation planning. A large number of metrics have been developed to estimate potential connectivity based on habitat configuration, matrix structure and information on organismal movement, and it is often assumed that metrics explain actual connectivity. Yet, validation of metrics is rare, particularly across entire landscapes undergoing habitat loss—a crucial problem that connectivity conservation aims to mitigate. We leveraged a landscape‐scale habitat loss and fragmentation experiment to assess the performance of commonly used patch‐ and landscape‐scale connectivity metrics against observed movement data, test whether incorporating information about the matrix improves connectivity metrics and examine the performance of metrics across a gradient of habitat loss. We tested whether 38 connectivity metrics predict movement at the patch (i.e. patch immigration rates) and landscape (i.e., total movements) scale for a pest insect, the cactus bug Chelinidea vittiger, across 15 replicate landscapes. Metrics varied widely in their ability to explain actual connectivity. At the patch scale, dPCflux, which describes the contribution of a patch to movement across the landscape independent of patch size, best explained immigration rates. At the landscape scale, total movements were best explained by a mesoscale metric that captures that distance between clusters of patches (i.e. modules). Incorporating the matrix did not necessarily improve the ability of metrics to predict actual connectivity. Across the habitat loss gradient, dPCflux was sensitive to habitat amount. Synthesis and applications. Our study provides a much‐needed evaluation of network connectivity metrics at the patch and landscape scales, emphasizing that accurate quantification of connectivity requires the incorporation, not only of habitat amount but also habitat configuration and information on dispersal capability of species. We suggest that variation in habitat may often be more critical for interpreting network connectivity than the matrix, and advise that connectivity metrics may be sensitive to habitat loss and should therefore be applied with caution to highly fragmented landscapes. Finally, we recommend that applications integrate mesoscale configuration of habitat into conn