Predictions of future population sizes are commonly based on estimates of population dynamical parameters such as the stochastic growth rate, the strength of density dependence, demographic variance and environmental variance as well as the uncertainty in the parameters. Measures of population size are often derived from imperfect data that contain observation errors, which can bias estimates of the population parameters. The environmental variance is a very vital parameter for predicting the distribution of future population sizes and it is thus important to separate environmental variance from observation error. If not properly accounted for, observation error will also bias other parameters such as the strength of density regulation and the stochastic growth rate.

Recently, the use of Markov Chain Monte Carlo (MCMC) techniques within a Bayesian state-space modelling framework has gained popularity for estimating parameters in presence of observation error. Based on studies of ibex in Swiss Alps, North American mallards (May Breeding Waterfowl Survey) and British birds (Common Bird Census) we will summarize our recent findings regarding advantages and potential pitfalls using these methods.

Results/Conclusions

In general, MCMC-methods are quite effective in separating observation error and environmental stochasticity, and estimates of environmental variance are unbiased even for short time series. Large uncertainty in estimates of environmental variance does however lead to wide population prediction intervals. This uncertainty can be reduced by longer time series or by reducing the magnitude of observation error. This implies that monitoring schemes could benefit from emphasising reduction of observation error, even at the expense of spatial or temporal extensiveness.

Estimates of the stochastic growth rate and the strength of density dependence are often biased such that the population growth at small population sizes is overestimated and return time to carrying capacity is underestimated. This can potentially cause to narrow prediction intervals as well as underestimates of the risk of extinction. The carrying capacity is however unbiased due to strong sampling correlation between the stochastic growth rate and the parameter describing density dependence.