Predicting the abundance of corals from simple environmental predictors with a machine-learning approach

Anderson B. Mayfield ； Alexandra C. Dempsey ； Chii-Shiarng Chen

《Platax》 2022卷 (2022/12) Pp. 1-24

https://doi.org/10.29926/platax.202212_19.0001

摘要

As Earth's oceans continue to warm at alarming rates, scientists have ramped up efforts to learn more about arguably the planet's most thermo-sensitive ecosystems: coral reefs. However, despite covering only a small areal fraction of the ocean, well under 1% of coral reefs have been surveyed, and many have likely never even been seen. For this reason, the Khaled bin Sultan Living Oceans Foundation (LOF) embarked on their "Global Reef Expedition" (GRE) from 2012 through 2016, characterizing thousands of never-before-studied reefs in all major coral reef areas and across a biological gradient that spanned molecules to ocean basins. We sought to leverage this rich dataset herein to identify areas of high coral cover that have not previously been surveyed, as this capacity could 1) aid in triaging conservation efforts and 2) reduce field time spent searching for "needle-in-a-haystack" reefs with exceptionally high coral abundance. We trained over 3,000 predictive models with various combinations of common environmental parameters (e.g., temperature, type of reef, etc.) from the LOF-GRE Solomon Islands dataset as a proof-of-concept, and one neural network featuring only nine of these predictors was associated with a validation R2 of 0.81. Although additional environmental and demographic predictors could be incorporated to attempt to more robustly predict the coral cover of unexplored reefs, this confidence is high enough to where managers and scientists could use the underlying model to predict where else in this Coral Triangle nation they are likely to find reefs with high abundance of live corals.

關鍵字

coral reefs ； global change biology ； machine-learning ； marine ecology ； Solomon Islands