Background/Question/Methods Persistent soil seed banks are recognized as a life history phenomenon that can buffer plant populations in variable environments. Representing seed banks in demographic models has proven more challenging: while aboveground individuals can be readily observed, additional experiments are often required to estimate seed survival, viability, and germination. Studies have identified consequences to omitting seed banks from stage-structured models and highlighted the effects of parameter uncertainty associated with dormant stages. We encountered these challenges when we sought to estimate seed survival and germination for population models of the winter annual Clarkia xantiana ssp. xantiana. We thus built on existing approaches by developing and evaluating a hierarchical model that includes a process component to represent loss from the seed bank via decay and germination, and a sampling model to represent the design of the seed burial experiments. Finally, we combined data from estimates of reproduction and seedling emergence in permanent plots with the seed experiments. We used our approach to assess the extent of intraspecific variation in the seed survivorship curve and in age-specific germination. Results/Conclusions We observed substantial variation in the survival of seeds in our C. xantiana populations, corresponding to Type I, II, and III seed survivorship trajectories. The Weibull survival process proved to be a good representation of persistence in the seed bank, and population-level estimates for the shape parameter ranged from .25 to 1.25. We summarized uncertainty about the full survivorship curve as a function of both the shape and inverse-rate parameter. We also propagated uncertainty about survival to estimates of age-specific germination. In addition to fitting our model to empirical data, we conducted a simulation study to aid with computational implementation. We compared a nonparametric representation of seed survival and several parametric survival functions (negative exponential, continuous exponential, Weibull). We simulated data from each survival process and fit models to confirm we recovered known parameter combinations. We also tested the consequences of fitting a model to data generated under a different process (e.g. fit a model with nonparametric survival to data generated by a Weibull). Overall, we propose that one of the strengths of the approach we describe is that it makes it possible to compare representations of seed decay and germination in the model fitting step before moving on to evaluating its impact on population growth rate. Our study contributes to efforts to make rich inferences from the trove of demographic data collected by plant ecologists.
