With a surface area of 68,000 sq. km (more than three times the size of Wales), Lake Victoria is the largest lake in Africa. Its near-circular shape, which has shores in Tanzania, Uganda and Kenya, makes it particularly vulnerable to intense thunderstorms.

"During the day, a breeze develops that flows from the cool water towards the warm land," said Wim Thiery, who is affiliated with both ETH Zürich, Switzerland and the Vrije Universiteit Brussel, Belgium. "At night, we see the opposite: the land breeze flows away from the cooling land towards the warmer lake. As the lake is shaped like a circle, these land breezes from all directions converge above the lake. Add evaporation to this cocktail and you get a lot of storms, rain, wind and waves."

Around 200,000 people fish in the productive waters of the lake, but the storms make fishing here a dangerous profession. The International Red Cross estimates that between 3000 and 5000 fishermen per year lose their lives in violent storms on Lake Victoria.

Several organisations are working on producing an early warning for those fishing on Lake Victoria. However, most of these forecasts are based on numerical weather prediction, and require running a model to forecast the weather and derive storm warnings from predicted atmospheric conditions. Thiery and his colleagues, in contrast, believe they have found a simpler and cheaper way to forecast these dangerous storms.

Studying NASA satellite observations gathered every 15 minutes between 2005 and 2013, the scientists noticed a strong statistical relationship between certain weather conditions and the probability of storms.

"We found that afternoon conditions on land influence how strong night-time storms on the lake will be, suggesting greater predictability than has been assumed so far," said Thiery. "Based on these new insights, we developed a statistical model, based on logistic regression, which evolved to a prototype of a new storm-prediction system called VIEWS."

By feeding current satellite observations into VIEWS, the team is able to generate storm warnings up to several hours in advance. In particular the model looks for the presence of overshooting tops – dome-like protrusions on cumulonimbus anvil clouds.

"Overshooting tops indicate the presence of deep convective events and are induced by intense updraughts through the tropopause into the lower stratosphere," Thiery and colleagues write in Environmental Research Letters (ERL) .

Now the researchers have made their experimental prediction system publicly available under the MIT licence. Ultimately, they hope to develop automatic warning messages that will be published on Twitter.

In addition, Thiery and colleagues are assessing whether their approach may be useful in other parts of the world, such as Lake Maracaibo in Venezuela, which experiences similar night-time thunderstorms to Lake Victoria and receives the most lightning strikes on Earth.

Related links

Related stories