How Pangu Weather Model helps protect fishing communities in Madagascar

(July 2024) Life in the villages near Fort-Dauphin in the southeast of the island of Madagascar looks idyllic. Palm trees, fishermen, barefoot children playing soccer, and no crime. The Ravinala, a symbol of Madagascar also known as Traveller’s Tree, grows freely. 

But first impressions are often misleading. Ever since people have settled in the area, sudden weather changes have surprised fishermen while they were at sea. Most residents have relatives who never came back to shore after unexpectedly encountering a storm.  Sometimes, the sea brings back pieces of the boat and some personal effects, and other times, nothing is found.

This threat from the sea appreciably started to lessen last year. That’s when the local NGO, Mitao Forecast, discovered Pangu Weather. 

In typical coastal villages in Madagascar, fishermen set out on small wooden canoes called lakanas. The frail vessels seem built for calm inland lakes and rivers, but Malagasies take them so far out to sea that the coast disappears on the horizon.  There, they become vulnerable to rogue waves and actual storms are a death sentence. 

Madagascar is a large island in the Indian Ocean off the south-eastern coast of Africa. It is among the countries most vulnerable to climate change. In recent years, cyclones have become fiercer and more sudden.

A fishing village near Fort-Dauphin

Until recently, no one was providing villagers with the localized weather data fishermen need to remain safe. The forecasts on TV are regional and of little practical relevance to specific coastal areas.  

This is a problem that Mitao Forecast identified. Founded by Toky Sylvestre in 2014, the NGO provides an early warning system that is a lifesaver for local fishermen. “I saw many children without fathers, orphans” Sylvestre recalls. “That’s when I began to think I need to take responsibility for a better solution.” 

Mitao Forecast’s mission is simple: warn fishermen when dangerous weather is on its way. The NGO uses data for different sources to generate local forecast. Every day, NGO volunteers post weather on billboards in fishing villages. Taking into accounts that many residents are illiterate, the forecasts display their message in green, yellow and red, with red indicating days when it’s not safe to go to sea.

In a village, Toky introduces the upgraded weatherboard with Pangu’s 10-day forecast.

To maintain credibility and be effective, Mitao Forecast needs to provide accurate forecasts. Last year, Toky and his team noticed Huawei Cloud’s Pangu-Weather Model when it became available on the website of the European Centre for Medium-Range Weather Forecasts (ECMWF). The site provides forecasts for anywhere the world. Mitao began relying on Pangu as another source to generate localized 10-day forecasts including wind speeds, temperature, atmospheric pressure and other data that Mitao’s meteorologists require. 

In January 2024, Tropical Cyclone Alvaro hit Madagascar. The storm would have been a catastrophe for any lakana at sea. But Mitao warned and prepared villagers of prolonged adverse weather conditions. Pangu was one of the key tools used to predict the severity and impact of this weather event. So far, Mitao’s forecasts reach 600,000 to 750,000 rural Malagasies. 

At dawn, villagers walk by a weatherboard.

Huawei Cloud's Pangu-Weather Model is the first AI model that can predict weather more accurately than state-of-the-art numerical weather prediction (NWP) methods, and at a speed that is 10,000 times faster. In the past, predicting the path of a typhoon for a 10-day period took 4 to 5 hours of simulation on a high-performance cluster of 3,000 servers. Now, Pangu-Weather can do that in 10 seconds on a single one-GPU server. Pangu-Weather can generate a global weather forecast within seconds, with all factors taken into consideration, such as geopotential, humidity, wind speed, temperature, and sea-level pressure. Besides typhoon paths, it can also accurately predict precipitation, cold and heat waves, and more.

On July 5, 2023, Nature, the world's top scientific journal, published a paper titled "Accurate medium-range global weather forecasting with 3D neural networks", which provided a detailed report on the research results of the Pangu-Weather model.

Since Mitao started to use Pangu to improve its forecasts, fishermen’s deaths at sea have become a small fraction of what they were before, Toky says. “My goal is to take technology like Pangu all over Madagascar. And after that, I want to take the Mitao Forecast to all of Africa and to the world.” The way he sees things, he has a responsibility to do this.