T201 The Future of Quantitative Paleontology: Biometry, Computer Vision, and Machine Learning

Tuesday, 6 November 2012
1:30 PM - 5:30 PM
217A

MODEL RANKING METHOD FOR BIOSTRATIGRAPHIC CORE ALIGNMENT

Paper number: T201-14
Program: Technical Sessions
Time: 5:00 PM – 5:15 PM

Authors

Abstract

We describe method to detect intervals of community stasis and change within core samples and align those cores across a transect to indentify the temporal and spatial extent of communities. We demonstrate our method using core samples from five sites within the Great Barrier Reef. Cores represent 1000 years of continuous accumulation of coral skeletons. Each core is segmented into samples and the data extracted from each sample is the relative biomass of 38 genera of corals. Each observation is a 38-dimensional vector of relative proportions that sum to one. As in other paleocommunity models, two observations are from the same community if they are plausibly drawn from the same probability distribution (all variations thus ascribed to sampling).

Our algorithm has two steps. The first step is to group observations within a core according to similar communities. Change points are detected statistically within each core using a model ranking method. We employ a dynamic programming algorithm to search for optimal segmentations using Akaike’s Information Criterion (AIC) and modify those results using Bayesian Information Criterion (BIC). Next, we use a modified version of Needleman Wunsch (NW) algorithm to align the segmented cores. The NW is known in bioinformatics as a method to compare and align sequences of macromolecules. We extend the scoring method used in the classical NW algorithm to use AIC to evaluate the statistical similarity of samples.

Results on coral communities show scientifically plausible grouping of samples and form the foundation of further studies on biodiversity and community stability.