Watch your time step: trapping and tracking dispersal in autocorrelated environments

1. Reliable estimates of dispersal kernels are needed for many applications in ecology. For both plants and animals, and for passive and active dispersal, the two most common methods for obtaining dispersal kernel estimates from direct measurements are following individual propagules (tracking) and counting propagules at fixed locations (trapping). 2. We consider the effect of the chosen sampling method on the estimation of dispersal kernels and highlight reasons why we should not necessarily expect estimates from the two methods to be the same. We illustrate this point using data from field experiments for three wind-dispersed plant species (Carduus nutans, Carduus acanthoides and Crepis praemorsa) and a mechanistic dispersal model (WALD). 3. The field experiments demonstrate that the two sampling methods lead to different dispersal kernel estimates. Although estimated mean dispersal distances may be similar, short-term tracking studies with random seed release generally lead to more peaked distributions with greater modal dispersal distances than long-term trapping of naturally dispersing seeds. 4. Wind speed data and dispersal distributions estimated using the mechanistic dispersal model suggest that differences between the sampling methods arise from temporally autocorrelated environmental conditions interacting with study time frame and dispersal initiation processes. Single tracking events tend to provide biased estimates of wind statistics because of short-term autocorrelation in wind speeds. Depending on the long-term autocorrelation structure of wind speeds, and the sampling of wind speeds compared to natural seed release, single short-term tracking studies may result in distributions that differ strongly from the 'true' distribution sampled by trapping over the dispersal season. Distributions from several tracking studies, weighted according to natural dispersal initiation, may therefore be needed to recapture the true dispersal distribution. 5. This study demonstrates that commonly used sampling methods for dispersal can result in dramatically different distributions under temporally autocorrelated environmental conditions. As autocorrelated environments are ubiquitous, this could lead to marked biases in predictions of movement and spread. These results are relevant for a wide range of ecological systems and to theoretical and applied problems relying on measurements and models of dispersal, for instance, conservation in fragmented landscapes, biological invasions and species re-introductions. The measurement process should be taken into account in such applications.