R goes fishing: Analyzing fisheries data using AD Model Builder and R
Arni Magnusson (University of Washington)
Fisheries stock assessment is mainly about two questions: what is the
current stock status, and what are the consequences of alternative harvest
levels. The data provide indirect information in the form of annual landed
catch, age distribution, index of relative abundance, and biological
features like length, weight, and maturity at age. Other possible data
include tag recoveries, juvenile surveys, and acoustic estimates of
absolute abundance. Due to these diverse data types, as well as economic
importance, fisheries stock assessment has pushed the limits of applied
statistics over the last fifty years. The dynamic models are highly
nonlinear, include dozens of parameters, and the objective function
consists of several likelihood components. The most efficient algorithm to
fit these models is automatic differentiation, as implemented with AD
Model Builder, a dialect of C++. To diagnose the model fit, trellis plots
in R have proven very useful, and the coda package is used to diagnose
MCMC convergence. Here, I demonstrate the functions I have implemented to
make these tasks easier, both for newcomers to R and old dogs.