The physics of the problem begins with the way seawater treats light. As a beam travels through water, two things happen to it. Photons are absorbed when water molecules convert optical energy into heat, a process captured by the wavelength-dependent absorption coefficient a(λ). Photons are also scattered when they collide with suspended particles — sediments, salt crystals, plankton, microscopic bubbles — redirecting them away from their intended path, described by the scattering coefficient b(λ). Together these define the total attenuation coefficient c(λ) = a(λ) + b(λ). Seawater is kindest to light in the blue-green band of roughly 450 to 570 nanometers, the celebrated “optical window” where attenuation reaches its minimum, which is why most underwater optical wireless communication (UOWC) systems have historically clustered there. Red and infrared light, by contrast, is absorbed by the vibrational resonances of water molecules, while ultraviolet is absorbed by dissolved organic matter.

The new study departs from convention by asking what happens when the link is not clean and direct. In practice, the receiver often collects photons that have been bounced into it by scattering — a scattered-light channel — because the transmitter and receiver are misaligned. The researchers modelled this explicitly: a 5-watt transmitter, a 10-meter link, a 60-degree receiver field of view, a silicon photodetector and a fixed misalignment of 45 degrees, an angle that suppresses the direct Lambertian beam component to a negligible level in coastal and turbid waters. Emission was described by a Lambertian intensity pattern, with the Lambertian order derived from a 30-degree half-power divergence angle, and the direct beam subject to Beer–Lambert exponential decay. For the scattered component, the team used a first-order model built around the scattering albedo ω(λ) = b(λ)/c(λ), the ratio that describes how likely an intercepted photon is to be redirected rather than absorbed. The authors are candid that this is a single-scattering approximation with a simplified phase function — adequate for clear and coastal water, but only indicative of trends in the murkiest conditions, where multiple-scattering paths become significant.

The framework’s central and arguably most novel move is what the authors call joint electro-optical optimisation. Previous wavelength studies in UOWC optimised on optical attenuation alone. This work folds in the wavelength-dependent responsivity of the silicon photodetector, R(λ), which rises steadily from blue toward the near-infrared. Because electrical signal current is the product of responsivity and received optical power, and signal-to-noise ratio scales as the square of both quantities, the detector’s red-biased sensitivity provides a multiplicative gain that shifts the system-level optimum away from the purely optical minimum-attenuation wavelength. It is this electro-optical coupling, the authors show, that explains a surprising and counterintuitive finding: in turbid water, the best wavelength is not blue-green at all, but red.

To test across ocean conditions, the team drew on the inherent optical properties dataset of Solonenko and Mobley, which indexes absorption and scattering coefficients for the standard Jerlov water classification. Three representative classes were simulated: Jerlov Type I, clear open-ocean water; Jerlov Type III, moderately turbid coastal water; and Jerlov Type 9C, highly turbid harbour and estuarine water. Sweeping the wavelength from 400 to 700 nanometers, the model predicted a clear-water optimum at 564 nanometers — comfortably inside the established blue-green window — with a signal-to-noise ratio of 26.03 dB, a Q-factor of 20, an effective bit error rate below the practical forward-error-correction floor of 10⁻¹², and channel capacity of 432.53 Mbps under the classical Shannon bound (372.3 Mbps under the more conservative intensity-modulation/direct-detection capacity bound).

The picture changes dramatically as turbidity increases. In Jerlov Type III coastal water, the model-predicted optimum migrates to the 650–700 nanometer red band, delivering 18.22 dB of SNR and roughly 304 Mbps of classical Shannon capacity. In Jerlov 9C turbid harbour water, the optimum stays in the same red band, but performance degrades further: 9.62 dB SNR, a Q-factor of 3.03, a bit error rate of 1.23 × 10⁻³ and about 167 Mbps of capacity. The red shift is not a quirk of the simulation boundary. Beyond 700 nanometers, the O–H vibrational overtone absorption of water rises steeply, physically capping the migration. The team therefore reports the turbid-water recommendation as a range rather than a single wavelength. The reason red wins in murky water is a three-way trade-off: short wavelengths scatter so strongly that photons are spread over a huge solid angle and mostly miss the receiver aperture; red light scatters far less, concentrating energy; and silicon detectors respond more strongly to red photons, multiplying the electrical signal even when optical power is modest.

Perhaps the most striking physical insight is that in coastal and turbid water, up to 90 percent of the received power arrives as scattered light rather than directly. This validates the entire premise of the study: communication is still possible when the direct path is essentially gone, because scattering itself becomes the channel. In clear water the direct component dominates at short ranges, and communication reaches approximately 18.2 meters before SNR falls below a 10 dB operating threshold; the corresponding ranges are about 13.0 meters for coastal water and 10.0 meters for turbid water.

The framework’s predictions stand up against independent published work surprisingly well. The clear-water optimum of 564 nanometers sits within 6.4 percent of the 530 nanometers used empirically in a pointing-adjustable beam array study of clear seawater, and the turbid-water recommendation of 650–700 nanometers overlaps almost perfectly with the 645-nanometer vertical-cavity surface-emitting lasers deployed in a harbour-like water demonstration. By contrast, studies that fix a single wavelength — such as one using 532 nanometers across all Jerlov water types — pay a measurable SNR penalty in coastal and turbid conditions, precisely the penalty this framework is designed to eliminate.

The authors also quantified how fragile these results are. A Monte Carlo error-propagation analysis with 200,000 samples revealed that clear-water links are highly robust, with an SNR standard deviation of just 0.65 dB and error rates never breaching the practical floor. Coastal water links were largely robust but breached the floor in roughly 28 percent of sampled parameter combinations. Turbid-water links were markedly less reliable: SNR varied by 5.55 dB, fifth-percentile SNR approached zero, and bit error rates exceeded the practical floor in over 91 percent of cases. Uncertainty in the scattering phase function matters most precisely where scattering dominates the signal — a statistically grounded confirmation that turbid-water results should be read as indicative trends pending more rigorous validation.

The practical implications reach beyond simulation. For engineers designing underwater sensor networks, autonomous vehicle coordination or defence surveillance links, the message is that wavelength should be treated as a design variable tied to environment. The authors outline a feasible adaptive architecture: a multi-chip transmitter combining blue-green and red emitters, switched electronically at the driver level without moving parts; water-characterisation via pilot probes at reference wavelengths or an in-situ transmissometer; and onboard evaluation of the closed-form link budget, which is fast enough to run in microseconds per wavelength on embedded processors. Because the optimal wavelength varies slowly with water type rather than with instantaneous channel fades, switching decisions need only be triggered by turbidity reclassification, not per-frame adaptation — a far gentler requirement than conventional adaptive modulation.

The team is explicit about what the work is and is not. It is a systematic, simulation-based analytical study, not an experimental demonstration or a new channel propagation model; its contribution is the integration of established component models — Beer–Lambert attenuation, Lambertian emission, first-order scattering, photodetector noise, on–off keying error rates and Shannon capacity — into a cohesive, wavelength-resolved NLOS optimisation framework. Future work includes Monte Carlo photon-tracing validation using full Henyey–Greenstein phase functions, an extension of the wavelength sweep into the near-infrared up to 850 nanometers, experimental tank-based validation, deep-learning-based channel estimation for real-time wavelength control, and a parametric sweep of misalignment angles from 0 to 90 degrees. Until those results arrive, the study’s headline recommendation stands as a provocative design principle: in the ocean, the best color for your beam depends on where you are — and a system smart enough to change color with the water can hold a link where a fixed one cannot.