AERIEL // MARINE SCIENCE // RESEARCH LOGS // RL-2091-033
SCIENCE ARM ONLINE ·
Research Log · Deep-Sea Photonics · RL-2091-033

Spectral Library of Abyssal Bioluminescence

An in-situ emission-spectrum library of abyssal bioluminescent taxa, built from low-light hyperspectral ROV imaging, to enable non-invasive species census and dark-navigation for the submersible fleet.

Deep-Sea Photonics Interim Method: in-situ hyperspectral ROV imaging

Abstract

Most abyssal animals make their own light, and most of what we know about that light comes from dying specimens hauled to the surface, where the chemistry is already wrong. We built a library of in-situ emission spectra — light measured where the animal lives, at depth, undisturbed — using a cooled low-light hyperspectral imager mounted on an ROV running dark. Across 47 dives between 800 and 3,200 m we captured calibrated spectra from 38 taxa. Emission clustered tightly around blue-green (peak 470–490 nm) as expected for transmission through deep water, but with diagnostic structure: several taxa showed secondary red emission and distinctive spectral “fingerprints” sufficient to classify them automatically. A spectral-match classifier identified known taxa at 91% accuracy from emission alone — meaning the fleet can census the deep without netting a single animal.

Background & Question

Two needs converged. The science arm wants a non-destructive way to survey abyssal communities (every trawl is damage). The fleet wants to navigate and identify in the dark without floodlights that blind and scatter the very animals they want to observe — and that betray a submersible's position. Bioluminescence answers both, if emission spectra are diagnostic enough to tell species apart. The question: do abyssal taxa emit spectra distinct enough to support automated identification in situ, and can a classifier do it at survey-useful accuracy?

Methods

Findings

The library holds calibrated emission spectra for 38 taxa. As predicted by deep-water transmission, peak emission concentrated at 470–490 nm, but the spectra were not interchangeable: bandwidth, secondary-peak structure, and decay kinetics separated taxa cleanly. The classifier reached 91% top-1 accuracy on held-out spectra, with confusions confined to closely-related ctenophores. Two taxa displayed a secondary red emission band (≥680 nm), a rare trait useful as a near-unique marker. The dataset is sufficient to deploy passive spectral census on the next survey series.

Taxa Catalogued38800–3,200 m
Classifier Accuracy91%spectral-only ID
Modal Peak478 nmblue-green
TaxonPeak (nm)FWHM (nm)Secondary peakDecay (ms)
Humpback Anglerfish (lure)48634
Vampire Squid (arm-tip)47241~900
Crystal Comb Jelly49052~120
Stomiid dragonfish48130705 nm
Siphonophore (colony)47538~600

Implications

A passive spectral census lets the League survey the deep without killing it — aligned with the doctrine that we damage as little as we must. The same library hardens the fleet: a submersible can identify what it is looking at, and navigate, on emission alone, without floodlights that reveal its position to a tracked surface vessel. Science and stealth share one dataset. The library is committed to the registry and cross-referenced to the abyssal holdings; spectra for unidentifiable emitters are flagged for the cryptid review queue, where one anomalous repeating pulse train remains unresolved.

References & Linked Records

Imaging note. Visual record pending generation. Art-direction prompt on file: “Pitch-black abyssal water lit only by living blue-green bioluminescence: a stomiid dragonfish and a comb jelly glowing against total darkness, faint ROV instrument glow at frame edge, hyperspectral scientific aesthetic, profound deep-sea silence.”
← Back to research logs