Spectral Unmixing and Mapping of Coral Reef Benthic Cover

Abstract

Coral reefs are important to the global ecosystem and the local communities and wildlife that rely on the habitat they create. However, coral reefs are also in critical and rapid decline: reefs have degraded over recent decades and what remains is at increasing risk of loss. Only a small fraction of the world's reefs have been studied quantitatively. Mapping these reefs, in particular the benthic cover components of coral, algae, and sand, would allow us to monitor changes and understand their health. We seek methods to better model reefs at global scales. Throughout this work, we train models on different levels of spatial-resolution satellite data and evaluate performance between them. We also develop on a novel method using a Deep Conditional Dirichlet Model to perform unmixing of spectral signatures in different satellite data sources. Accounting for mixing of classes yields improvements in accuracy of the predicted class probabilities. Regions of high uncertainty are also determined based on the model output prediction and can be used to determine informative in situ sampling. An ergodic planner is implemented to generate a path through these regions to acquire samples that best improve the coral map. The result of this study is an efficient learning based pipeline that augments existing satellite data and maps coral reefs globally in order to improve our understanding of their condition.