UW Aquatic & Fishery Sciences Quantitative Seminar
A general approximation to time-varying parameters in ecological models
Ecological models generally approximate biological and sampling processes with simple equations whose parameters are constant over time. However, many processes will vary randomly or systematically over time, causing the value of modeled parameters to also change. In this talk, I present a generalized approximation to time-varying parameters, which uses stepwise model building and the Akaike Information Criterion to select an appropriate smoothness for fixed-knot splines that approximate time-varying parameters. This stepwise-spline approximation does not require integration (unlike random effects) and can be conducted simultaneously with other model building considerations. It also generalizes block-effects, random walks, linear interpolation, and many other common approximations to time-varying effects. It is demonstrated with two case studies: a multi-state, robust-design tag-resighting model for South African sea turtles, and an occupancy model for Australian elasmobranchs. These case studies demonstrate the improvements in accuracy and interpretability that can be gained by using relatively simple approximations to time-varying parameters in ecological models.