Models have
become standard tools when assessing the state of fish stocks, evaluating
recovery strategies, and in making sustainable harvest decisions. They
typically consist of several interrelated equations to describe complex
biological and ecological processes. These equations are expressed in terms of
unknown variables (parameters) that collectively define the model parameter
set.
This
parameter set is estimated through computational procedures that use uncertain
observations. A procedure (algorithm) is considered efficient when it is stable
(i.e., given the same input information, it returns identical estimates of
parameters), fast, and precise. Model predictions or model-based inference that
use parameters from these efficient algorithms are considered reliable and
optimal. The ADMB/TMB platforms currently used for
fisheries modelling and stock assessment are built on such an algorithm.
This study
uses simple illustrative examples to demonstrate how inconsistent parameter
estimates may result from an algorithm that is considered fast and precise.
Ignoring such inconsistencies may result in non-optimal parameter estimates,
wrongly calibrated models, and erroneous model-based decisions. For fish stock
assessment models, this may lead to erroneous inference about population size,
and wrong estimates of parameters that are central to management. As
consequence, management decisions may result in non-sustainable fisheries or
misguided recovery plans.
Though these
issues are central to modelling in fisheries science, discussion on them has
been limited to scientists with a strongly quantitative background. This paper
adopts language that allows for a more inclusive discussion.