AD Model Builder is a package designed to meet the requirements posed by typical stock assessment models (nonlinear, highly parametrized, possibly time-varying parameters). Published benchmarks have shown that it provides faster and more reliable parameter estimation than other generic function minimizers. This is achieved with automatic differentiation (AD) and the programming interface is a thin layer on top of C++, with convenient features to read and write data files, perform vector and matrix calculations, with optional features like random effects and MCMC analysis. Model input and output is in plain text files, that can be analyzed and plotted in R or other statistical packages. AD Model Builder is free software (http://admb-project.org), originally written by Dave Fournier, the 2009 recipient of the American Fisheries Society’s William E. Ricker Award.
After going through biomass-dynamic models, statistical age-structured models and MCMC analysis, the focus will be on random effects and finally a State-space Assessment Model (SAM), which is used for several assessments in ICES. This is a full stochastic model that allows selectivity to vary gradually with time, and can handle years with missing data. It has fewer model parameters than full parametric statistical assessment models, as quantities such as fishing mortalities and stock sizes are modelled as random effects.