| Abstract Detail
Biogeography Salinas, Nelson R. [1], Wheeler, Ward C. [1]. Statistical Modeling of Areas of Endemism: a Markov Random Field approach. A statistical framework to infer areas of endemism from geographic distributions is proposed. This novel method is based on Hidden Markov Random Fields, a type of undirected graph model commonly used in computer vision. This framework assumes areas of endemism are the states of the hidden layer of the model, whereas taxon distributions are emitted values in the observed layer. Taxon distributions are associated to the observed layer through a clustering procedure based on the extent of overlap. Observations are emitted by the hidden layer according to a Gaussian distribution, whereas the joint distribution of the hidden layer follows a Potts model. State and parameter inference of the maximum a posteriori configuration is performed through a modified version of the Expectation-Maximization algorithm. The optimal number of areas of endemism in the dataset is estimated through the Pseudolikelihood Information Criterion, a model selection procedure that uses an approximation to likelihood. The performance of the new algorithm was assessed on simulated data, and compared to the most popular methods for delimitation of areas of endemism: Biotic Element Analysis, Parsimony Analysis of Endemism, and Endemicity Analysis. Hidden Markov Random Fields efficiently recovered the true pattern across a wide range of uncertainty values. Additional analyses on empirical data were conducted on a dataset of Neotropical Blueberries (Ericaceae: Vaccinieae) distributed mainly on the Andean Cordillera.
Log in to add this item to your schedule
1 - American Museum of Natural History, Division of Invertebrate Zoology, Central Park West at 79th street, New York City, New York, 10024-5192, United States
Keywords:
areas of endemism
geographic distributions
Hidden Markov Random Fields
biogeography.
Presentation Type: Oral Paper
Session: 13, Biogeography
Location: Sundance 3/Omni Hotel
Date: Monday, June 26th, 2017
Time: 2:00 PM
Number: 13003
Abstract ID:142
Candidate for Awards:None |
