Predicting the spread of invasive species in an uncertain world: accommodating multiple vectors and gaps in temporal and spatial data for Bythotrephes longimanus

Real-world uncertainties and data limitations make it difficult to predict how, when and where non-indigenous species (NIS) will spread. Typically only a small fraction of sites are sampled during only a few time intervals, such that we know neither the full spatial extent nor the true temporal progress of invasion. Yet, these unsampled locations might affect the invasion dynamics. We extend propagule pressure models to incorporate both human-mediated and natural fluvial dispersal vectors, and develop techniques to incorporate missing spatial and temporal data on invasions. We apply our model to Bythotrephes longimanus, a high-risk aquatic NIS, using a regional-scale 311-lake survey in a popular watershed in Ontario and extending our analysis to 1,300 unsampled lakes. Of 100 model runs with different random subsets of 50 sampled lakes reserved for validation, we were able to obtain an average area under the curve value of 0.89. Human-mediated dispersal accounted for 99.75% of the contribution of propagules to probability of establishment. Although the discovery rate is accelerating, our results suggest the annual rate of lake invasions is decelerating over time. Management efforts controlling recreational boating traffic out of the largest lakes in the system will be the most effective way of slowing the spread of B. longimanus in lakes within this system.