OOS 17-10
Using time series of population densities and size-structures to infer population dynamics: Exploring feasibility scenarios

Tuesday, August 11, 2015: 11:10 AM
317, Baltimore Convention Center
Edgar J. Gonzalez, Mathematics and Statistics, McMaster University, Hamilton, ON, Canada
Carlos Martorell, Ecologia y Recursos Naturales, Universidad Nacional Autonoma de Mexico, Mexico, DF, Mexico
Benjamin M. Bolker, Mathematics and Statistics, McMaster University, Hamilton, ON, Canada

The study of the population dynamics of size-structured species has traditionally required the follow-up of individuals to determine their survival, growth and fecundity rates. With these rates, a directly estimated projection model can be used to predict the change over time of the population’s size structure and density. Apart from the logistical difficulties of repeated measures, this approach cannot be applied to the study of long-term directional changes through time or to species that are hard to track over time. An inverse method would instead use as input the time series of size structures and densities to estimate the vital rates that produced the series. However, survival, growth and fecundity are confounded in such data. Therefore, different rates could produce the same time series, making prior information on the life history of the species necessary. Nonetheless, the question still remains of how much information is required to obtain sensible solutions, and how this depends on sample size and/or environmental noise. This question was addressed within a Bayesian framework using a size- and time-varying integral projection model. Through simulations, different scenarios of data availability and environmental noise were explored.


Preliminary results show that a composite likelihood where population densities have a larger weight than the data on size-structures better restricts the space of possible solutions. As expected, in the absence of prior information on the biology of the species, the model fails to provide biologically sensible solutions, regardless of sample size. Conversely, the use of strongly informative priors guarantees sensible solutions, but the solutions are highly dependent on the priors. Increasing sample size allows for weaker priors as well as stronger environmental noise in the data. This approach is promising for long-term data sets where populations are periodically surveyed (but only at the level of their size-structure and density) and, to a lesser extent, for chronosequence data with large amounts of environmental variation or endangered species where sample size is intrinsically limited.