Calibrar: an R package for fitting complex ecological models

1. The fitting or parameter estimation of complex ecological models is a challenging optimization task, with a notable lack of tools for fitting complex stochastic models. 2. calibrar is an R package that has been developed for fitting complex ecological models to data, including complex Individual Based Models. It is a generic tool that can be used for any type of model, especially those with non-differentiable objective functions. 3. calibrar supports multiple phases and constrained optimization. It implements maximum likelihood estimation methods and automated construction of the objective function from simulated model outputs. 4. User-level expertise in R is necessary to handle calibration experiments with calibrar, but there is no need to modify the model's code, which can be programmed in any language. For more experienced users, calibrar allows the implementation of user-defined objective functions. 5. The package source code is fully accessible and can be installed directly from CRAN. Introduction The ability to achieve accurate parameter estimation has been used as a criterion to assess the usefulness of ecological models (Bartell 2003). Given a model, the criterion for the selection of the best possible parameter set is the optimization of a scalar objective function (e.g. log-likelihood, residual sum of squares) with respect to the model parameters (Walter and Pronzato 1997, Bolker et al. 2013). Once the objective function is properly defined, parameter estimation is essentially an optimization problem. The parameter estimation or calibration of complex ecological models (Jorgensen XXX) could be a difficult task for optimization because of model characteristics such as non-linearity, high dimensionality as well as low quantity and quality of observed data (Tashkova et al. 2012). These diverse factors have hampered the development of calibration algorithms and methodologies for ecological models that are sufficiently flexible and generic, and only sparse documentation has been produced on fitting complex models (Bolker et al. 2013). Additionally, complex ecosystem models can be numerically intensive and require long simulation runs, adding an extra layer of difficulty to their fitting. There are some dedicated tools for non-linear parameter estimation, AD Model Builder (ADMB; Fournier et al. 2012) being one of the most robust and fast (Bolker et al. 2013). Among other advantages, ADMB provides support for parameter estimation in multiple phases (Nash and Walker-Smith 1987), which can be of great interest when dealing with complex ecosystem models (Oliveros-Ramos et al. 2015). It also provides support for constraining optimization, which can be helpful …