Modelling unbiased dispersal kernels over continuous space by accounting for spatial heterogeneity in marking and observation efforts

Abstract

Abstract Although a key demographic trait determining the spatial dynamics of wild populations, dispersal is notoriously difficult to estimate in the field. Indeed, dispersal distances obtained from the monitoring of marked individuals typically lead to biased estimations of dispersal kernels as a consequence of (i) restricted spatial scale of the study areas compared to species potential dispersal and (ii) heterogeneity in marking and observation efforts and therefore in detection probability across space. Here, we propose a novel method to circumvent these issues that does not require data on observation effort per se, to correct for the variability in detection of marked individuals across space. Observed dispersal events were weighted by the distribution of departure points and an eroded spatial window approach was applied so as to deal with border effect. We conducted a set of simulations which indicated that our method was successful in correcting the effect of spatially heterogeneous detectability and produce unbiased dispersal kernels. We applied this method to a real dataset on the Montagu's harrier (>5,000 chicks tagged), providing c. 6,000 resightings collected in entire France by a network of 1,200 volunteers within a citizen‐science programme. The median dispersal distance observed was 32 km (range: 0.1–627 km). Once corrected for spatial heterogeneity in marking and observation efforts and border effect, the modelled dispersal kernel indicated a median dispersal distance of 78–123 km depending on the spatial scale considered (constrained within French borders or not, respectively). Synthesis and applications: The current rise of citizen‐science programmes is likely to stretch our estimate of the ecologically relevant spatial scale at which dispersal takes place for many taxa. Our method is particularly suited for such large‐scale data that typically suffer from high spatial heterogeneity in marking and observation efforts and offers the possibility to derive unbiased dispersal kernels, a key component for modelling population dynamics and species distribution in a context of environmental change. Currently, our method assumes homogeneity in both habitat and dispersal behaviour across individuals. We discuss, however, how to relax these hypotheses to further investigate the effect of, e.g. local conspecific density or habitat quality on dispersal propensity.