Human pressure, habitat heterogeneity and species traits jointly determine the level of dark diversity in coastal fish communities

Understanding how anthropogenic pressures and environmental gradients shape coastal fish communities is a central question in biogeography and marine conservation. Yet, the set of species expected to occur at a site given ecological conditions but being absent – the so‐called dark diversity – has received far less attention than observed species, even if they hold key information about local extirpations and conservation effectiveness. Quantifying dark diversity in the sea is particularly challenging because many species are rare, cryptobenthic or mobile, while sampling effort is rarely standardized. Here we combined 930 environmental DNA (eDNA) samples collected at 401 sites along 2000 km of the French Mediterranean coast with fine‐scale mapping of habitats, anthropogenic pressures and protection levels. Using a Bayesian model of dark diversity affinity (DDA) that explicitly controls for sampling effort, we decomposed the correlates of species absences into site characteristics and species traits, separating dispersal‐related from sensitivity‐related signals. At the site level, fishing effort, marine traffic, and artificial habitats were associated with greater DDA, so a deficit of species, whereas protected areas and diverse natural habitats were associated with lower DDA, increasing the likelihood of observing more complete fish communities. At the species level, demersal and predator species, together with those bearing their eggs (e.g. many elasmobranchs) compared to species with demersal or pelagic eggs, exhibited the highest DDA, suggesting local extirpations linked to a combination of limited dispersal and higher sensitivity to cumulative pressures. Beyond the creation of marine protected areas, preserving habitat heterogeneity and mitigating coastal pressures therefore emerge as essential levers to limit the local erosion of fish biodiversity. More broadly, coupling eDNA with a sampling‐effort‐corrected DDA framework offers a transferable benchmark to evaluate where, and for which species, conservation efforts are most needed.