Developing demographic models to inform selection of Alliaria petiolata (garlic mustard) biological control agents

Biological control is often considered a safe and effective method for controlling invasive plant species. While methods are available for predicting biocontrol agent host specificity, biocontrol practitioners currently lack effective tools for predicting agent efficacy. Demographic models which account for spatial and temporal variation in population dynamics promise to improve the predictability of weed biological control programs, while lowering the risks they pose to non-target species. Alliaria petiolata (garlic mustard) is an obligate biennial forb that is invasive in North American forests. I analyzed sources of demographic variation in twelve unmanaged A. petiolata populations in Michigan and Illinois, USA, and over three plant generations. These data were used to parameterize matrix population models of A. petiolata population dynamics, analyze A. petiolata responses to simulated management, and inform the selection of effective biological control agents for potential release in North America. Hierarchical, generalized linear mixed models (GLMMs) were used to analyze the spatial and temporal structure of variability in each demographic transition. The degree of variation observed in A. petiolata demographic rates was greater than expected based on previous studies of this species. This variation was highly structured in space and time and exhibited negative density dependence and positive response to precipitation across most of the life cycle. Estimates of the population growth rate (lambda) ranged from 0.48 to 5.88 across all sites and years. Within sites lambda was temporally variable, ranging from 0.80 to 5.88 within one site. A megamatrix model was used to summarize variation in growth within sites. Site growth rates (lambda M) ranged from 0.83 to 3.54 (mean = 1.90). Sensitivity and elasticity analyses of matrix population models indicated the importance of the seed bank to A. petiolata's success. Sensitivity and elasticity rankings varied with lambda, indicating that the transitions with the largest impacts on population growth differ for growing and declining populations, and within populations during good and bad years, rendering management options a moving target. Rosette survival (summer and winter) consistently emerged as the transition with the greatest effects on lambda in populations with positive growth, as did germination of new seeds and transitions affecting fecundity. This result is consistent with past predictions that rosettes should be targeted by management. The model raises the caveat that rosette survival is only a an effective target when growth is positive; its proportional effect on lambda decreases as lambda decreases. These models predict a lower probability of suppressing A. petiolata with biocontrol than past studies. The simulations predict that the root-crown weevil Ceutorhynchus scrobicollis could control up to 5 of 12 populations if introduced alone. Introducing a second species could extend control to as many 9 populations, although the probability of success is very low (< 0.1) at 4 of these 9 sites. Better data on the distribution of agent impacts are necessary to refine these predictions. Variance in survival was negatively density dependent, even when mean survival was not significant. By modeling the residual variance in each vital rate as a function of density, demographic variance and stochasticity themselves become density dependent functions. This has potentially important consequences for populations of management concern, as small populations may become more susceptible to local extinctions. The predictive power of future weed management models may be improved by incorporating density dependent demographic stochasticity in their designs.