Evaluation of stochastic gravity model selection for use in estimating non-indigenous species dispersal and establishment

Abstract

Predicting dispersal of nonindigenous species (NIS) is an essential component of risk analysis and management, as preventative measures are most readily applied at this stage of the invasion sequence. Gravity models provide one of the most useful techniques available to model dispersal of nonindigenous invasive species (NIS) across heterogeneous landscapes, as these models are able to capture transport patterns of recreational boaters who are dominant vectors responsible for aquatic NIS dispersal. Despite the widespread use of gravity models in forecasting biological invasions, different classes of gravity models have not been evaluated regarding their comparative ability to capture recreational transport patterns and subsequent use in predicting NIS establishment. Here we evaluate model selection between unconstrained, total-flowconstrained, production-constrained and doubly-constrained stochastic gravity models to assess dispersal of the spiny waterflea Bythotrephes between Ontario lakes. Differences between the models relate to the amount of data required and constraints under which calculations of source/destination interactions are made. For each class of gravity model, we then estimated the probability of a lake having established Bythotrephes populations by modeling the relationship between empirical presence/absence data and inbound recreational traffic (i.e. propagule pressure) via boosted regression. The unconstrained gravity model provided the best fit to observed traffic patterns of recreational boaters. However, when output from the gravity models was used to predict Bythotrephes establishment, the doubly-constrained gravity model provided the strongest relationship between inbound recreational traffic and observed Bythotrephes presence/absence, followed by the production-constrained model. Our results indicate production-constrained gravity models offer an acceptable balance between modeling recreational boater traffic and their utility to estimate establishment probabilities.

The spread, establishment and impacts of the spiny water flea, Bythotrephes longimanus, in temperate North America: a synopsis of the special issue

“…Research in this collection dramatically enriches this understanding. In independent modeling efforts, Potapov et al (2011);, and Muirhead and MacIsaac (2011) all demonstrate the central role of propagule pressure in explaining the current pattern of Bythotrephes presence on the Canadian Shield. Further prove that the component of propagule pressure contributed by stream connections in this landscape is so low it can be practically ignored, while it certainly can be high in much larger rivers (e.g.…”

“…Such comparisons are rarely possible, but are very useful for consideration of the consequences of subtle differences in model structure (i.e., analysis of model uncertainty), for identification of the potential importance of different underlying invasion processes, and for testing alternative hypotheses when multiple processes or model structures yield similar fits to the data. We assemble that research here, with four papers focused on modeling the growth and spread of the invader on the south- Muirhead and MacIsaac 2011;Potapov et al 2011;Wang and Jackson 2011). Combined, this work provides insight into where potential colonists are going, how quickly they are moving, and which sites will allow them to survive and prosper, information crucial to understanding and managing secondary spread.…”

“…Bythotrephes incidence is increasing in lakes on the south-central Canadian Shield (Weisz and Yan 2010), and the modeling efforts of Potapov et al (2011), Muirhead and MacIsaac (2011) and in this collection were built on that growing data set. However, Kerfoot et al (2011) add their own survey data to other recent American survey data (e.g.…”

“…There are methodological advances, regarding the maximal usage of incomplete spatial and temporal information , and the influence of the underlying structure of gravity models on their predictive ability. Production-contrained gravity models may well be the best overall choice (Muirhead and MacIsaac 2011). More fundamentally we learn that the ongoing invasion of CAISN's key 1600-lake watershed is actually slowing, despite increased discovery rates, likely because of saturation of optimal sites (Gertzen and Leung 2011).…”

Increased abundance of the non-indigenous zooplanktivore, Bythotrephes longimanus, is strongly correlated with greater spring prey availability in Canadian Shield lakes

“…Bythotrephes longimanus is a Palaearctic cladoceran that was first detected in the Great Lakes in the early 1980s (Johannsson et al 1991) and has since invaded [140 North American lakes (Branstrator et al 2006;Weisz and Yan 2010;Muirhead and MacIsaac 2011). The invader is a voracious and active predator that consumes a large number of zooplankton each day, with a preference for herbivorous cladocerans (Vanderploeg et al 1993;Burkhardt and Lehman 1994;Wahlström and Westman 1999;Grigorovich et al 1998).…”