Density dependence is a central concept in population dynamics. Fitting ‘mechanistic’ models to data is often used to detect density dependence and provide information on maximum reproductive rates and carrying capacity. However, choosing among the many possible mechanistic models is often difficult given the available data and estimates of population parameters may be highly model dependent. We introduce conditional Gaussian processes for use as prior distributions to account for the uncertainty in the shape of density dependence. We validate the method using simulated data and then apply the method to population time series for a wide range of species.

Results/Conclusions

Our semiparametric Bayesian method can recover underlying mechanistic models and hence can detect density dependence. The amount of confidence in the estimated model is related to the amount of data and the associated noise level over a given interval. Our method provides estimates of maximum reproductive rates and carrying capacity that agree with true values when calculated from simulated data. This approach avoids the common pitfall of model misspecification by using data to infer the functional form of density dependence and is robust in dealing with the ever-present uncertainty in ecological data.