Technical Note: Approximate Bayesian parameterization of a process-based tropical forest model

[摘要] Inverse parameter estimation of process-based models is a long-standingproblem in many scientific disciplines. A key question for inverse parameterestimation is how to define the metric that quantifies how well modelpredictions fit to the data. This metric can be expressed by general cost orobjective functions, but statistical inversion methods require a particularmetric, the probability of observing the data given the model parameters,known as the likelihood.

For technical and computational reasons, likelihoods for process-basedstochastic models are usually based on general assumptions about variabilityin the observed data, and not on the stochasticity generated by the model.Only in recent years have new methods become available that allow thegeneration of likelihoods directly from stochastic simulations. Previousapplications of these approximate Bayesian methods have concentrated onrelatively simple models. Here, we report on the application of asimulation-based likelihood approximation for FORMIND, a parameter-richindividual-based model of tropical forest dynamics.

We show that approximate Bayesian inference, based on a parametric likelihoodapproximation placed in a conventional Markov chain Monte Carlo (MCMC)sampler, performs well in retrieving known parameter values from virtualinventory data generated by the forest model. We analyze the results of theparameter estimation, examine its sensitivity to the choice and aggregationof model outputs and observed data (summary statistics), and demonstrate theapplication of this method by fitting the FORMIND model to field data from anEcuadorian tropical forest. Finally, we discuss how this approach differsfrom approximate Bayesian computation (ABC), another method commonly used togenerate simulation-based likelihood approximations.

Our results demonstrate that simulation-based inference, which offersconsiderable conceptual advantages over more traditional methods for inverseparameter estimation, can be successfully applied to process-based models ofhigh complexity. The methodology is particularly suitable for heterogeneousand complex data structures and can easily be adjusted to other model types,including most stochastic population and individual-based models. Our studytherefore provides a blueprint for a fairly general approach to parameterestimation of stochastic process-based models.

[学科分类] 地球化学与岩石