Optical Ocean Recipes: Methods & Applications

• Optical Ocean Recipes are a set of physically grounded procedures that combine calibrated additives or sensor configurations to emulate real seawater optical properties.
• They employ rigorous methodologies like non-negative least squares fitting and real-time calibration to achieve reproducible attenuation, scattering, and background light measurements.
• The framework spans controlled tank experiments, deep-ocean observatories, image correction for abyssal mapping, and adaptive remote sensing, emphasizing structured and measurable optical workflows.

Searching arXiv for papers on "Optical Ocean Recipes" and related ocean optics contexts. “Optical Ocean Recipes” denotes a family of physically grounded procedures for constructing, measuring, or interpreting ocean-related optical conditions across several distinct research settings. In the most explicit usage, the term refers to a framework for creating realistic, repeatable, and ground-truthed underwater image datasets in a controlled tank environment by mixing calibrated color and scattering additives to approximate measured seawater absorption and scattering spectra (Schöntag et al., 24 Sep 2025). More broadly, the phrase also captures a set of operational strategies in deep-ocean observatories, ocean-colour remote sensing, atom-optical tracer analysis, and even reflected-light studies of ocean-bearing exoplanets, where optical signals are treated as structured observables rather than incidental by-products (Henningsen et al., 2022). Across these domains, the common logic is to decompose optical phenomena into measurable ingredients—attenuation, scattering, radiance, photon counts, directional structure, background light, and environmental context—and then combine them into repeatable observational workflows.

1. Controlled optical water classes and dataset construction

The most literal formulation of Optical Ocean Recipes is the tank-based framework introduced for underwater vision research. In that framework, a “recipe” is a specification of how much of each calibrated optical additive should be mixed into a known volume of clear tank water so that the resulting water approximates a chosen target ocean water type in terms of absorption and scattering spectra (Schöntag et al., 24 Sep 2025). The targets in the reported experiments are measured Jerlov water types IB, II, and 3C, using spectral measurements reported by Williamson and Hollins. The additives are three food colorants—Patent Blue VF (E131), Copper-Chlorophyllin (E141), and Brown HT (E155)—and one scattering agent, magnesium hydroxide, Mg(OH)$_2$ (Schöntag et al., 24 Sep 2025).

The framework is explicitly tied to standard underwater image-formation physics. Its starting point is the Lambert–Beer law,

$I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$

with total attenuation

$c(\lambda) = a(\lambda) + b(\lambda),$

and with constituent additivity written as

$$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$

(Schöntag et al., 24 Sep 2025). This additive formulation is the physical basis for recipe computation: the target absorption is fitted by non-negative least squares over the measured base absorption spectra of the colorants, and scattering is then completed with magnesium hydroxide so that the total attenuation reaches the reference spectrum (Schöntag et al., 24 Sep 2025).

The tank workflow is designed for reproducibility rather than ad hoc visual plausibility. All base solutions and final mixtures are spectrally measured with a HITACHI U2800 UV-VIS transmission spectrophotometer. A “virtual standard solution” is defined as one whose attenuation spectrum integrates to $1\,\text{mm}^{-1}$ over the visible range from 400 to 700 nm, so actual base solutions can be normalized despite small preparation differences (Schöntag et al., 24 Sep 2025). The tank volume is 1600 liters, and optical scaling is used to compensate for the short physical path lengths: the attenuation coefficients are scaled by five so that, for example, a true distance of 0.5 m appears optically like 2.5 m (Schöntag et al., 24 Sep 2025).

The reported recipes are concrete. Under a scaling factor of five, Jerlov IB uses 221 ml Patent Blue VF, 25 ml Copper-Chlorophyllin, 15 ml Brown HT, and 4.4 g Mg(OH)$_2$; Jerlov II uses 221 ml Patent Blue VF, 39 ml Copper-Chlorophyllin, 33 ml Brown HT, and 4.4 g Mg(OH)$_2$; Jerlov 3C uses 234 ml Patent Blue VF, 115 ml Copper-Chlorophyllin, 163 ml Brown HT, and 21.0 g Mg(OH)$_2$ (Schöntag et al., 24 Sep 2025). The paper argues that these mixtures are spectrally closer to real seawater than improvised substitutes such as black tea, chamomile tea, milk, Maaloxan, aluminum hydroxide, and a green bath product (Schöntag et al., 24 Sep 2025).

This controlled-recipe interpretation of ocean optics is significant because it supplies realistic but controllable physical water conditions, with measurable optical properties and repeatable imaging. A plausible implication is that it occupies a methodological middle ground between synthetic rendering, which provides perfect ground truth but may suffer a realism gap, and open-water testing, which is operationally realistic but typically lacks controlled optics and ground truth (Schöntag et al., 24 Sep 2025).

2. Deep-ocean observatories as optical water-column recipes

A second, very different meaning of Optical Ocean Recipes emerges in deep-ocean observatory design. The Pacific Ocean Neutrino Experiment, P-ONE, is a proposed cubic-kilometer-scale neutrino telescope in the northeast Pacific, on the Juan de Fuca Plate in the Cascadia Basin, roughly 300 miles west of Vancouver Island, at a depth of around $2660\,\mathrm{m}$ (Henningsen et al., 2022). The observatory is to be built in collaboration with Ocean Networks Canada’s optical-electrical cabled observatory system, so the site is conceived as a permanently powered, networked deep-sea platform with communications to shore rather than an isolated instrument (Henningsen et al., 2022).

The detector concept most relevant to ocean observations is a distributed three-dimensional array of moored observatories. In the current concept, one cluster contains 10 mooring lines, each $1\,\mathrm{km}$ long, rising upward from the seafloor at $I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$0. Each mooring line carries 20 instruments used either for light collection or calibration, is connected to a mini junction box at the seafloor, and distributes power and communications through an electric-optical backbone cable (Henningsen et al., 2022). The optical modules contain multiple photomultiplier tubes capable of single-photon detection with nanosecond timing resolution (Henningsen et al., 2022).

Within this architecture, the primary physics purpose is to detect Cherenkov light from neutrino-induced particle interactions, but the same PMT array also measures ambient optical conditions in the deep ocean (Henningsen et al., 2022). The STRAW and STRAW-b pathfinders, deployed in 2018 and 2020, were explicitly designed to characterize the Cascadia Basin in terms of optical properties and optical background as a function of time (Henningsen et al., 2022). STRAW used $I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$1 mooring lines and STRAW-b used $I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$2 mooring lines (Henningsen et al., 2022). According to the P-ONE letter, these systems showed that the site is optically clear enough for a large neutrino telescope and revealed that long-term monitoring of bioluminescence is itself a compelling scientific use case (Henningsen et al., 2022).

In this observatory setting, the “recipe” is architectural rather than chemical. The four ingredients named in the P-ONE letter are a permanently powered and networked deep-sea site in optically clear water; a vertically and horizontally distributed array of extremely sensitive optical detectors; colocated environmental sensors including temperature, accelerometers, pressure, acoustic positioning, and potentially cameras; and a sustained operational mode with shore communications, calibration capability, and interdisciplinary access to sensors and data products (Henningsen et al., 2022). This suggests an observational program in which photon-counting PMTs characterize ambient optical background and bioluminescence continuously, while acoustic and mechanical positioning reconstruct line motion and currents, and temperature-related measurements supply thermodynamic context (Henningsen et al., 2022).

The strongest oceanographic emphasis in this observatory conception is on bioluminescence. Because P-ONE is highly light-sensitive, it can continuously collect photons from the ambient deep-ocean environment, not just from particle events (Henningsen et al., 2022). The included STRAW-b example is a five-minute record of PMT count rates from two photosensor instruments, with separate up- and down-facing PMTs, and the substructure in these rates is interpreted as varying bioluminescence activity through the sampled water column (Henningsen et al., 2022). A plausible implication is that the detector can function as a depth-resolved optical background observatory at abyssal depth, where background light is dominated by intrinsic and biological sources rather than solar illumination.

3. Site characterization, attenuation, and real-time calibration in dynamic deep water

The STRAW site-characterization study provides the most quantitative deep-water optical recipe for Cascadia Basin. STRAW consists of two $I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$3 mooring lines separated horizontally by $I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$4, deployed at $I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$5 depth, instrumented with three calibrated light emitters and five receiver modules (Bailly et al., 2021). The effective wavelengths used are 365, 400, 450, and 585 nm (Bailly et al., 2021). After two years of operation, the reported attenuation lengths are

$I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$6

$I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$7

$I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$8

$I = I_0 e^{-(a(\lambda)+b(\lambda)) \cdot x},$9

(Bailly et al., 2021). The study defines the direct-light relation as

$c(\lambda) = a(\lambda) + b(\lambda),$0

with $c(\lambda) = a(\lambda) + b(\lambda),$1 combining absorption and scattering for the prompt component (Bailly et al., 2021).

The attenuation analysis is based on the hit fraction per flash, modeled as the probability of at least one detected photon,

$c(\lambda) = a(\lambda) + b(\lambda),$2

with

$c(\lambda) = a(\lambda) + b(\lambda),$3

(Bailly et al., 2021). Timing histograms of $c(\lambda) = a(\lambda) + b(\lambda),$4 are used to isolate prompt signal peaks above a uniform random background. A 50 ns integration window is adopted because strong-signal peaks have FWHM of 8–14 ns, while weak signals can broaden to 50 ns (Bailly et al., 2021). This indicates a measurement strategy in which geometry, source anisotropy, angular acceptance, and efficiency priors are jointly fitted with the attenuation length.

The same experiment also characterizes ambient background. Using 3-inch PMTs, the two-year single-PMT rate distribution has a 10th percentile of $c(\lambda) = a(\lambda) + b(\lambda),$5, a median of $c(\lambda) = a(\lambda) + b(\lambda),$6, and a 90th percentile of $c(\lambda) = a(\lambda) + b(\lambda),$7, with occasional excursions above $c(\lambda) = a(\lambda) + b(\lambda),$8 (Bailly et al., 2021). The background shows a periodic modulation of about 12.5 hours matching the tidal cycle and no significant seasonal variation (Bailly et al., 2021). $c(\lambda) = a(\lambda) + b(\lambda),$9 plus dark noise provide a persistent floor, while strongly time-variable bioluminescence dominates the dynamic background (Bailly et al., 2021).

A complementary real-time calibration recipe is developed for the TRIDENT Pathfinder, T-REX, in the western Pacific at $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$0 depth (Tian et al., 2024). That study introduces a custom monochromatic CMOS camera system, with 5 million pixels and a 25 mm focal length lens, mounted in deep-sea light receiver modules and paired with a central light emission module operating at 405, 460, and 525 nm in steady mode (Tian et al., 2024). Two receiver baselines, $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$1 and $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$2, are used (Tian et al., 2024). Within 30 minutes, about 3000 images were captured (Tian et al., 2024).

The rapid attenuation estimator in that system uses mean gray values near the light-source centroid in the two cameras: $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$3 so that

$$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$4

(Tian et al., 2024). Full radial image profiles are then compared with Geant4 simulations to infer $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$5 and $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$6 separately (Tian et al., 2024). The paper emphasizes that deep ocean water is dynamic and potentially non-uniform, so a distributed, real-time calibration network is preferable to occasional local measurements (Tian et al., 2024). This suggests a shift from static site survey to continual optical state estimation inside large detector volumes.

4. Image correction and mapping on the dark seafloor

Optical Ocean Recipes also describes a practical image-preprocessing methodology for abyssal visual mapping, where no surface light reaches and robots carry co-moving light sources through total darkness (Köser et al., 2021). In this regime, the same 3D point appears with different color in each image because illumination geometry changes with vehicle motion, while scattering and attenuation make images foggy and typically blueish (Köser et al., 2021). The paper models the observed intensity as

$$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$7

where $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$8 is seafloor appearance, $$c(\lambda) = \sum^i \left( a_i(\lambda) + b_i(\lambda) \right)$$9 is a multiplicative illumination/attenuation field, and $1\,\text{mm}^{-1}$0 is additive backscatter (Köser et al., 2021).

The method is tailored to predominantly homogeneous, flat seafloor regions such as abyssal plains (Köser et al., 2021). Its first step is to estimate the additive backscatter image from water-column frames acquired before reaching the bottom: $1\,\text{mm}^{-1}$1 The second step is to construct a virtual all-seafloor image from a short temporal window over mapping imagery,

$1\,\text{mm}^{-1}$2

and then estimate the multiplicative field up to a scale by subtracting the additive component and dividing by an assumed sediment reference color (Köser et al., 2021). The corrected image is then

$1\,\text{mm}^{-1}$3

The paper characterizes the method as essentially parameter-free in the sense that it avoids explicit models for lamps, camera, water, or scene, and does not require large annotated training sets (Köser et al., 2021). In practice, however, it still uses concrete implementation choices: a spatial median, downsampling of undistorted 12 MP images by a factor of 8, a temporal median on 7 images, and subsequent upsampling of the estimated factor image (Köser et al., 2021). With this strategy, the CUDA implementation estimates $1\,\text{mm}^{-1}$4 and outputs corrected 12 MP images at about 2 Hz, which is faster than the 1 Hz acquisition rate (Köser et al., 2021).

This mapping recipe is not intended to recover absolute physically calibrated radiance or absolute spectral albedo. Rather, it produces scatter-reduced, illumination-compensated, overlap-consistent imagery, and, if a sediment reference color is chosen, a relative-albedo-like estimate up to a global white-balance factor (Köser et al., 2021). The restriction to flat, homogeneous seabed is explicit. If real scene variations occupy more than half the temporal or spatial support, the median-based procedure can force true albedo changes into the estimated illumination field (Köser et al., 2021). The method is therefore best understood as a robust operational substitute for full physical restoration in a narrow but important abyssal-imaging regime.

5. Ocean-colour remote sensing and coastal bio-optical retrieval

In satellite ocean colour, Optical Ocean Recipes refers to operational band-based workflows that convert top-of-atmosphere radiance into coastal bio-optical products while recognizing that open-ocean assumptions often fail in sediment-laden and optically complex waters (Gupta et al., 2024). For EOS-06 OCM-3, the total radiance at the sensor is decomposed as

$1\,\text{mm}^{-1}$5

where the terms respectively denote Rayleigh scattering, aerosol scattering, Rayleigh-aerosol interaction, sun glint, whitecaps, and water-leaving radiance (Gupta et al., 2024). Sun glint is written

$1\,\text{mm}^{-1}$6

(Gupta et al., 2024).

The practical difficulty is that the standard NIR black-pixel approximation works very well over open ocean but fails over sediment-laden and optically complex waters, where water-leaving radiance in the NIR is no longer negligible (Gupta et al., 2024). EOS-06 OCM-3 addresses this with 13 VNIR bands spanning approximately 400–1010 nm, including 780, 870, and 1010 nm bands used for atmospheric correction (Gupta et al., 2024). The operational recommendation is explicit: use 780 and 870 nm over open ocean, and use 870 and 1010 nm over optically complex or turbid waters (Gupta et al., 2024).

After atmospheric correction, the paper provides empirical band-ratio formulas for chlorophyll-$1\,\text{mm}^{-1}$7 and $1\,\text{mm}^{-1}$8. For chlorophyll,

$1\,\text{mm}^{-1}$9

with

$_2$0

(Gupta et al., 2024). For $_2$1,

$_2$2

with

$_2$3

(Gupta et al., 2024). The paper reports $_2$4 and RMSE $_2$5 for the log-transformed chlorophyll fit, and $_2$6 with RMSE $_2$7 for the $_2$8 fit (Gupta et al., 2024).

This remote-sensing recipe is empirical rather than semi-analytical. It operates primarily on apparent optical properties such as $_2$9 and $_2$0, and it does not provide an explicit inversion for inherent optical properties such as absorption $_2$1 or backscattering $_2$2 (Gupta et al., 2024). Its strength lies in connecting sensor spectral design to practical atmospheric-correction branch points and product formulas. A plausible implication is that, in coastal waters, “recipe” should be understood as adaptive band selection plus uncertainty propagation rather than a single globally fixed atmospheric correction.

6. Atom-optical tracers, broader interpretations, and analogical extensions

A distinct but related usage appears in ocean ventilation studies that solve an oceanographic tracer problem using a laser-based atom-counting technique. In the $_2$3Ar dating study, ocean ventilation is diagnosed using Atom Trap Trace Analysis specialized as ArTTA, which measures the chemically inert noble-gas radioisotope $_2$4Ar with half-life $_2$5 years and atmospheric isotopic abundance $_2$6 (Ebser et al., 2018). The dating relation for a simple parcel is

$_2$7

with

$_2$8

(Ebser et al., 2018). Because ocean waters are mixtures, the study instead interprets measurements with a transit time distribution,

$_2$9

(Ebser et al., 2018).

The optical part of the recipe is the atom-optical apparatus. Metastable argon is produced in an RF discharge, laser-cooled and focused, slowed in a $_2$0 m Zeeman slower, and trapped in a magneto-optical trap, where single trapped $_2$1Ar atoms are detected by fluorescence on an avalanche photodiode and CCD (Ebser et al., 2018). The practical breakthrough is sample size: historical low-level counting required around 1000 L of water, whereas ArTTA uses only 5 L of water, corresponding to about $_2$2 STP of argon needed for analysis (Ebser et al., 2018). This is an optical recipe in the literal sense of “small water sample $_2$3 argon extraction $_2$4 laser trapping of rare atoms $_2$5 isotope counting $_2$6 ocean ventilation diagnosis” (Ebser et al., 2018).

The phrase can also be extended metaphorically beyond terrestrial oceans. In the exoplanet study of ocean planets, the operational problem is to identify which broadband optical features in reflected total flux and polarization phase curves uniquely reveal a surface liquid ocean (Trees et al., 2019). The reflected light is represented by the Stokes vector

$_2$7

with degree of linear polarization

$_2$8

(Trees et al., 2019). The paper concludes that the color change in polarized flux $_2$9 with increasing phase angle above $2660\,\mathrm{m}$0 appears to uniquely identify an exo-ocean, independent of surface pressure or cloud fraction, while a color change in $2660\,\mathrm{m}$1 identifies a partly cloudy exo-ocean (Trees et al., 2019). This is not oceanography in the usual marine sense, but it preserves the same recipe structure: isolate the optical observable, decompose the controlling physics, and identify a diagnostic combination that is robust to nuisance variability.

Taken together, these usages define Optical Ocean Recipes as a cross-domain methodology rather than a single instrument class. In tank datasets, the recipe is a calibrated mixture matched to target water spectra; in deep-sea observatories, it is a distributed sensor architecture that continuously measures attenuation, background, and environmental covariates; in visual mapping, it is a robust decomposition into additive and multiplicative optical nuisances; in ocean-colour remote sensing, it is an adaptive band-selection and inversion workflow; in tracer studies, it is a laser-based atom-counting chain; and in exoplanet optics, it is a reflected-light diagnostic for ocean-bearing worlds (Schöntag et al., 24 Sep 2025). The shared principle is that ocean optics becomes most informative when expressed as a reproducible sequence of physical ingredients, measurement steps, and inferential constraints.