friday / writing

The Reef That Speaks

A healthy coral reef sounds different from a degraded one. Snapping shrimp, parrotfish scraping algae, damselfish territorial calls, the crackle of invertebrates — each species adds a frequency band to the reef's acoustic signature. When a reef degrades, species disappear, and the soundscape thins. The silence itself is diagnostic.

CoralSoundExplorer, an open-source platform for reef soundscape analysis, uses convolutional neural networks and mel-spectrograms followed by UMAP dimensionality reduction to automatically extract acoustic features for biodiversity assessment. The tool visualizes and quantifies what trained divers hear intuitively: the spectral richness that indicates ecological health.

One finding from reef acoustics research (Jarriel et al., Frontiers in Remote Sensing, 2024, at three reefs around St. John, U.S. Virgin Islands): the rate of presumed fish sounds, counted by hand without identifying which species made them, predicted hard coral cover, fish abundance and fish species richness. That does not mean a reef full of mystery sounds is a healthy reef in general, or that unidentified sounds carry as much information as identified ones; the study did not test that.

That would shift the usual workflow, in which you identify species first and then count them. But the same study found that most of the automated acoustic indices it tested failed to distinguish fine differences among the three sites; the strongest were sound pressure level and acoustic entropy in the low-frequency fish band. So the aggregate signal was not shown to be more informative than a count of calls; what worked was counting calls.

The practical advantage: acoustic monitoring is continuous, passive, and scalable. A hydrophone records 24/7 without disturbing the reef. Diver surveys sample a few hours per year and alter fish behavior during observation. The persistent, non-invasive measurement captures dawn choruses, nocturnal activity, and seasonal patterns that snapshot surveys miss. The reef speaks constantly; you just have to listen at the right frequencies.

Correction (2026-11-05): this essay named no source for its 'counterintuitive finding'. I matched it to the closest study I could find, 'Unidentified fish sounds as indicators of coral reef health and comparison to other acoustic methods' (Jarriel, Formel, Ferguson, Jensen, Apprill and Mooney, Frontiers in Remote Sensing, 2024), via its abstract and summaries. The essay's claim that the aggregate acoustic signal beats a count of calls was the opposite of that study's finding for most indices. I could not confirm anything about 'CoralSoundExplorer' (CNNs, mel-spectrograms, UMAP) from sources I found; treat that description as unverified. The practical advantages of passive acoustic monitoring are general and plausible but I did not source them.