3D surface analysis of coral microatolls

Abstract

Microatolls are coral colonies with living outer margins but flat, dead upper surfaces. Coral microatolls are also valuable paleo-sea-level indicators owing to both their vertical precision and the fact that they can grow and respond to sea level changes for decades to centuries, unlike other short-lived organisms. In addition, they also can be used to track tectonic changes associated with their area of growth.

Studying microatolls is typically done by slabbing, an invasive process where a radial slice is obtained in order to study its annual growth bands. This is an invasive method. In this paper, we aim to lay a foundation for non-invasive analysis of coral properties by studying the 3d surface of the corals collected through LIDAR and photogrammetry techniques. The coral mesh is broken down into patches, and surface patch similarity is measured by comparing their corresponding embeddings, generated via a convolutional mesh autoencoder, which employs hexagonal convolutions in its routine. In addition, we perform detailed mathematical mesh analysis with ideas of incorporating smoother features obtained as a result into our autoencoder in the future.

The results obtained are mostly human interpretable and lay the foundation for use of machine learning methodologies in studying coral surfaces. In addition, a multithreaded framework for speeding up the preprocessing needed by the mesh autoencoder is also presented.