Measuring long-distance seed dispersal in complex natural environments: an evaluation and integration of classical and genetic methods

Abstract

Summary 1. Seed dispersal is a critical life stage of plants, yet accurate measurement of dispersal distances has been difficult in natural systems. Genetic techniques for matching dispersed seeds to maternal trees provide valuable data on dispersal events. Questions remain regarding how best to estimate the population seed dispersal distance distributions from such data and how these estimates compare with classical non-genetic estimates based on seed trap data alone. 2. Using simulated data, we compared seed shadow estimates obtained via standard inverse modelling of seed arrival into seed traps within mapped stands (summed seed shadow, SSS) with estimates from four models using genetic matches: direct fitting of the observed distribution of distances in the genotyped sample (observed distance, OBS), direct fitting inversely weighted by sampling intensity (OBSw), inverse modelling of numbers of seeds of each genotype in each trap (gene shadow model, GSM), and inverse modelling of frequencies of each genotype in each trap (competing sources model, CSM). We further explored how consideration of immigrant seed rain from unmapped and ungenotyped trees outside the stand affected dispersal estimates, denoting these variants SSSi, GSMi and CSMi. 3. We applied these models to an empirical data set for the Neotropical tree Jacaranda copaia, using a hierarchical Bayesian model to incorporate variation in fecundity among trees. 4. Fits to simulated data sets showed that OBS and SSS estimates were strongly biased, while SSSi, GSMi and CSMi were mildly biased. Root mean square errors varied as OBS >> SSS > OBSw > CSMi > GSMi > SSSi > CSM > GSM. 5. Comparing results for Jacaranda under the three best models, mean posterior estimates of dispersal distances varied as SSSi < GSM < CSM, but credible intervals overlapped among all three models, demonstrating agreement that long-distance dispersal is common. 6. Synthesis. Here we reconcile two general approaches used to study seed dispersal distances in natural communities. Genetic and non-genetic approaches can both provide good estimates of seed dispersal provided that analyses of genetic data take account of any deviation from random selection of seeds for genotyping, and that SSS models consider immigrant seed rain in the analysis whenever there is reason to think its contribution is non-zero. The use of the models presented here should provide better estimates of population-level dispersal distance distributions in genetic and classical seed dispersal studies.