Fitting these models to large datasets often requires colossal amounts of computational power and may give inaccurate results. Accounting for these aspects is essential when modeling the behavior of marine animals, but results in complex models. In addition, the ocean is not static and models should disentangle ocean drift from the voluntary movement of animals. For example, accurate positioning systems are not well suited to the marine environment and models must account for the effect of measurement error. Using such an approach with marine data is challenging. Tackling the Challenges of Fitting Movement Models to Marine Data Įcologists often investigate animal behavior by applying statistical models to movement data. The R package Template Model Builder (TMB) combines automatic differentiation with the Laplace approximation for general model building in a Maximum Likelihood framework. These difficulties only multiply when we try to find the (observed) Fisher information. In hierarchical models the joint distribution of observed and unobserved data is easy to describe, but the marginal distribution of observed data is often not.Įven in fully observed models finding the score functions can be difficult and time consuming. Maximum Likelihood Estimation consists of three steps: Describing the probability of the data, solving the score equations to find an estimate, and using the Fisher information to describe the variance of the estimator. Maximum Likelihood Estimation Using Template Model Builder Organizer and Chair: Joanna Mills Flemming (Dalhousie University)ĬHRISTOFFER ALBERTSEN, Technical University of Denmark Using TMB to Quickly and Robustly Solve Problems from Marine Ecology
0 Comments
Leave a Reply. |