Invasive species mapping from aerial pollen counts and occurrence survey data

Background/Question/Methods

Even as fossil pollen analysis has become an essential tool for inferring long-term vegetation changes at local, regional or continental scales, there are few examples of the use of contemporary pollen records, specifically those from aerial samplers, to better understand the dynamics of extant plant populations. Despite their limited spatial resolution and temporal extent, aerobiological samples may be useful in monitoring fast-changing populations of invasive plants. Since they constitute a continuous record in time that depends on plant abundance, pollen concentration measurements can complement occurrence data obtained from punctual field surveys. To estimate abundance maps from these two types of data, we propose a hierarchical Bayesian modelling approach where both pollen concentrations and field observations depend on a latent dynamic process, representing the evolution of the plant's population density in space and time. We test this method by modelling the distribution of common ragweed (Ambrosia artemisiifolia) in central and southeastern France, using pollen data from the Réseau national de surveillance aérobiologique (RNSA) and occurrence data from a recent synthesis of Ambrosia observations in Europe. We use the STAN and JAGS packages to implement these hierarchical models in R.

Results/Conclusions

Using pollen concentrations from 29 RNSA stations covering the 2000-2011 period as well as presence survey data from 2011, we estimate the parameters of a diffusion-growth process for ragweed densities, discretized on a 40x40km grid defined over the 520x520km study area. The survey data, even though it was produced by opportunistic sampling, improves the spatial interpolation accuracy of the model for areas poorly covered by aerobiological stations: based on cross-validation results, this accuracy also exceeds that obtained by ordinary kriging of the pollen data. However, reliable estimates of the diffusion parameter would require additional occurrence data near the beginning of the pollen time series.