The effects of climate change on coral reefs are leading to longer periods of ocean warming, which, in turn, is increasing the occurrence and severity of coral bleaching globally. These events, coupled with the growing data streams from monitoring bleaching incidents, are creating a bottleneck in the data processing required for reporting.
To address this issue, we have developed a workflow to replicate in situ bleaching severity assessment which relies on observing the bleaching intensity in individual coral colonies during and after events. We propose combining automated point classification with semantic segmentation for colony-level bleaching estimation, aiming to minimize disruptions to existing point count analysis methods. The workflow is model agnostic, allowing users to apply either foundational or bespoke segmentation models, and their existing point classifiers.
We evaluate this workflow using bleaching events and time series surveys on the Great Barrier Reef in 2024 and 2025. Our results demonstrate that machine learning can supplement the bleaching assessment pipeline with an accuracy level comparable to that of humans while achieving a 70% efficiency gain.
This research illustrates that machine learning can enhance the bleaching severity analysis from benthic image surveys, providing greater speed in reporting without compromising accuracy. Furthermore, this approach can be applied post-processing to coral point count image annotations.