The accelerating loss and fragmentation of natural habitats are major threats to biodiversity, the very foundation of our life-supporting system. High land pressure at the global scale reduces opportunities to preserve large amounts of habitat. Conserving habitat connectivity, that relies on organisms’ easiness to move between habitat patches, thus appears as the only possible option to sustain biodiversity. But setting up priority targets to preserve or enhance connectivity is not straightforward as it requires to estimate and parameterize the fluxes of organisms among habitat patches. To date, most connectivity conservation studies use dispersal kernels for this purpose. We combine individual-based modelling and laboratory-based movement observation to address this challenge of parameterising the movement of individuals to flows of individuals in schematic landscapes. We show that usual approaches over-estimate connectivity, but also that individual and contextual variability matter. We also show that landscapes with spatially heterogeneous resistance to movement can lead to counter-intuitive movement fluxes. Constant back and forth between observations and models and between models is mandatory to deal with the many problems of scale encountered in ecology and ecological modelling.