Despite hundreds of hours in the water and dedicated efforts by 20+ scientists, we canvassed less than 1% of Palau’s coral reefs; is it possible to use knowledge on these reefs to predict what the 99% we did NOT survey may look like? Could we leverage the data to project where those reefs with the highest coral cover may be? This is the subject of this article based primarily on these data.
By having trained a machine-learning algorithm in JMP Pro, we can at least provide you with some clues. The model below is associated with a validation data R2 of ~0.7. By plugging in known coordinates and reef conditions, you can get an estimate of what the coral cover and mean colony resilience may be.
We have also shown the results of a desirability analysis, in which the AI was commanded to simulate the conditions associated with the highest coral cover. Those looking for the most coral-rich habitats, then, should consider reefs with the following properties:
Here is another neural network modeling the characteristics of resilient pocilloporid corals in Palau (Mayfield & Dempsey, 2025). In this figure, the Y in panel b is the probability of a coral being resilient (“p(resilient)”).
A relatively “simple” neural network for predicting whether a coral would demonstrate resilience or not (“sick”) based on assessment of 22 environmental and ecological (namely benthic survey) parameters, followed by a desirability analysis showing conditions predicted to be associated with a resilient coral at 99% certainty.
Coral health index (CHI) plotted across the entirety of Palau’s western (leeward) side, as well as plots depicting absence of correlation between CHI and coral abundance (from Mayfield & Dempsey, 2025).