Submarius
The water clarity forecast.
Published April 16, 2026 · Last updated September 2, 2026
Submarius estimates horizontal underwater visibility, in feet or meters, for any coastal point. It starts from that location's typical clarity for the time of year, built from in water measurements and satellite ocean color, then subtracts what changes day to day: wave stirring judged against depth and bottom type, river and runoff plumes, and tide driven water movement. Each estimate shows a low to high range, its data sources, and how fresh they are.
Most marine apps use surface chop as a stand in for visibility, which fails in bays, sounds, and estuaries. Submarius forecasts the visibility itself.
How the forecast works.
The visibility number is the output of an ocean-optics model, not a lookup table. Five stages, built to be credible from day one and to sharpen over time:
Multiple satellite ocean-color sources (NOAA CoastWatch VIIRS and Copernicus) are combined into an estimate of the water's inherent optical properties: how much it absorbs and scatters light. Where multi-sensor data isn't available, we derive them from Kd490 and chlorophyll (a Kirk decomposition).
A radiative-transfer forward model (Mobley + Duntley) turns those optical properties into real horizontal diver visibility, and caps it at the depth, because you can't see past the seafloor.
Physical models for wave- and tide-driven sediment resuspension, river-discharge plumes, rainfall runoff, and harmful algal blooms pull visibility down, or let it recover.
A calibration layer scores the model against public in water measurements (the Water Quality Portal, AERONET-OC ocean color sites) and refits its priors weekly. A bounded machine learning correction, trained on those real measurements and constrained by the physics, removes systematic error. Divers' one tap post dive reports are collected to validate and improve the model over time.
Every estimate is a range (p10/p50/p90), not a single fake-precise number. Spots unlike anything we've calibrated on get wider bands, and where we have no direct data we say so instead of inventing a number.
What the drawer actually shows you.
Tap the Visibility card in the app and you drop into an immersive underwater scene: a dogtooth tuna swimming at 20 feet, a diver silhouette at 40 feet, both attenuated by Beer-Lambert optics at the current vis level. The fish fades. The diver fades harder. You can see what the number means, not just read it.
Below the scene: the visibility estimate, confidence percentage, reason, and a 24-hour forecast trend. No hype. No "WATER IS GREAT!" buttons.
Honesty over hype.
A cornerstone principle at Submarius is the data honesty rule: we never claim what we don't measure. If we're using a proxy (like swell height as a "surface state" hint), we label it as a proxy. If we don't have direct turbidity data for your region yet (many inland lakes, remote coastlines), we tell you that, not invent a number.
For every dive-mode verdict, the dialog includes an always-on caveat: "We don't yet have direct water-clarity data for this location. Apply local knowledge." A user who's burned by a "GOOD" verdict on a blown-out day loses trust forever. Trust is the product.
Yes, there's machine learning in it, but on a short leash: a bounded correction on top of a physics model, trained on real in-water measurements, with every input and the uncertainty always shown. Not an "AI" black box that invents a number.
"Best vis window this week."
Applying the same model to forecasted future feature vectors gives you the three hour stretch in the next seven days with the highest predicted clarity. Essential for anyone planning a weekend trip.
Private even from us.
Community vis reports contribute to the model, but your precise dive location never crosses the wire. Only an H3-tile-fuzzed (~1 km) coordinate is submitted with the report. Personal dive logs (with exact GPS) live in device storage only. If a Submarius employee queried the database tomorrow, they'd see who dove in what general area, not your spots.
Common questions.
We don't read it off a single satellite number. The forecast starts from each location's typical clarity for the time of year, built from in water measurements and multiple satellite ocean color sources (NOAA CoastWatch and Copernicus), then subtracts what changes day to day: wave stirring judged against depth and bottom type, river and runoff plumes, and tide driven water movement. Every estimate shows a low to high range and the data source behind it.
Most marine weather apps use surface only signals (wave height, swell) as a proxy for visibility. That breaks down in estuaries, bays, and coastlines where the water can be turbid regardless of how flat the surface is. A real clarity forecast requires satellite ocean color data, river plume tracking, and geometry aware reasoning, work that hasn't historically been done outside scientific research.
It's built on published ocean optics and checked against public in water measurements (the EPA Water Quality Portal and AERONET-OC ocean color sites). The Lee 2015 relationship the optics draw on reports R² = 0.96 in the reference literature; that is the published theory, not our field accuracy. We are preparing a public calibration page with our own measured error by region, and where the model has no direct data for a spot the app says so instead of inventing a number.
Evidence and related reading.
- Weekly water clarity reports — the model's regional reads, published every Tuesday.
- How to read diving conditions — the plain English guide to the signals.
- Ocean conditions glossary — 77 terms, from Secchi depth to slack tide.
- How to predict water visibility and the Kd490 deep dive.
- Public methodology docs — data source citations and feature explainers.
See how clear today actually is.
Want to measure it yourself?
We are designing an affordable clarity sensor you could anchor at your own spot, from open ocean to an inland spring. Tell us what it would need to do.
Take the 3 minute survey