For divers in the UK and Ireland, predicting the underwater visibility at dive-sites at any given moment can present a major challenge.
Changeable weather is already a factor, but nobody wants to undertake a time-consuming and costly trip to the coast only for it to be wasted on sediment-packed underwater views. Conditions can remain a mystery until the first divers have dipped their heads beneath the surface.
Now a freediver has developed an AI tool that he says can predict the vis at specific locations to enable more dependable dive-planning. Eventually Dima Karamshuk hopes to be able to extend the reach of his Marla Blue app, and to map water quality with ever-greater degrees of accuracy across the world’s oceans.
Karamshuk’s career to date has revolved around machine-learning. His further education in Ukraine and Italy culminated in a PhD in computer science, after which he became a post-doctoral researcher at King’s College, London, working on BBC iPlayer and other projects; principal staff scientist at online flight-booker Skyscanner; and, for the past six years, a staff research scientist for Meta.
He is now based in Devon, having moved there from London to pursue his passion for freediving. Diving regularly off the coast, however, he found that the variable and seemingly unpredictable underwater visibility was affecting his enjoyment, and often making his dives feel less than safe.
“I got tired of having dives with poor vis, so I thought there must be a way to figure it out,” he says. With many other divers complaining of the same problem, especially on social media, in early 2023 he decided to apply his artificial intelligence expertise to the problem.
This led him to co-found Marla Blue, combining 10 years of satellite observations with offshore and onshore wind, swell, rain, chlorophyll-concentration, wave and topography data to create a forecasting algorithm.
“We’ve been getting good results,” he says. “We’ve done hundreds of experiments, used years and years of coastal data and hundreds of diver reports to tweak, refine and improve our model.”
The Marla Blue app is now into its V3.1 incarnation. The name? “I was playing around with la mar in Spanish; a marine-scientist friend also read MARine LAboratory into it, but that’s probably a stretch!” says Karamshuk.
Rising accuracy
But the important question is this: How well does the app work? Last summer the team started asking divers to report on the visibility encountered on their dives, so that their experiences could be analysed and compared with Marla Blue forecasts.
Over the course of the season, they received some 500 reports, and 58% of these were found to be within the predicted ranges of the current V3.1 model.
This was encouraging progress, because earlier models had been in the 43-47% range, and it represented a 29% improvement over the previous live model, V2.2.
The diver reports came in from all over the UK and Ireland but especially the South-west (287) and North-east (121) regions. The most enthusiastic response came from Beadnell Point in Northumberland, thanks to one devotee named Scuba Steve, and some reports were submitted from inland waters.
Most of the best visibility was recorded in the South-west and Ireland, though the best of all, a sparkling 20m, came from near St Abbs in south-eastern Scotland in May.
Average vis was 4.12m with the (probably more representative) median 3.5m and variability of 2.86m.
True to human nature, Marla Blue users tend to report more often when frustrated by app predictions that have proved incorrect, so this has to be taken into account. “We’re confident that we’ll be able to increase the model accuracy,” says Karamshuk.
Global expansion
The focus dive-sites are split into three categories: Shore, Reef and Wreck, and app-users can obtain underwater visibility forecasts not only for specific locations but for anywhere around the coast.
The forecasts are updated hourly, with visibility estimates provided for today, tomorrow and the past three days, along with algae trends for the past week and site conditions including detailed wind, swell, tide and water-temperature data, along with local diver reports.
Marla Blue, which trades as Ocean AI, secured its initial funding from Innovate UK in April 2023, with another tranche from the same source and from Maritime Launchpad in the Great South West the following year.
With an eye to the mentioned global expansion, Karamschuk is currently experimenting with dive-visibility forecasts in Australia. “Forecasting undersea visibility is tough, but we’re up for the challenge,” he says. “With every divers’ valuable input, we believe in our collective ability to map the undersea conditions across the world’s oceans.”
While Marla Blue is already supported by a growing community of scuba and freedivers, its founder is currently looking to further boost its development by encouraging all divers, whether app-users or not, to complete a short survey covering aspects such as problems encountered when planning dives, and preferred types of graphic.
The Marla Blue site provides further information and offers a useful guide to finding good diving conditions in the UK and Ireland. The app is available from the IoS and Android stores, priced at £35 a year – commencing with a one-week free trial – or £6 a month.
