🔗 Share this article

When Tropical Storm Melissa was churning off the coast of Haiti, meteorologist Philippe Papin felt certain it was about to escalate to a major tropical system.

Serving as lead forecaster on duty, he predicted that in a single day the storm would intensify into a severe hurricane and begin a turn in the direction of the coast of Jamaica. Not a single expert had previously made such a bold forecast for rapid strengthening.

However, Papin had an ace up his sleeve: artificial intelligence in the form of Google’s new DeepMind cyclone prediction system – released for the first time in June. True to the forecast, Melissa did become a storm of remarkable power that ravaged Jamaica.

Increasing Dependence on Artificial Intelligence Predictions

Forecasters are increasingly leaning hard on the AI system. During 25 October, Papin explained in his public discussion that Google’s model was a primary reason for his certainty: “Approximately 40/50 Google DeepMind ensemble members indicate Melissa becoming a Category 5 hurricane. While I am not ready to predict that intensity yet given path variability, that is still plausible.

“It appears likely that a phase of rapid intensification is expected as the system drifts over very warm ocean waters which represent the highest oceanic heat content in the whole Atlantic basin.”

Outperforming Traditional Systems

Google DeepMind is the pioneer artificial intelligence system dedicated to tropical cyclones, and now the first to outperform traditional meteorological experts at their specialty. Across all 13 Atlantic storms this season, Google’s model is the best – surpassing experts on path forecasts.

The hurricane ultimately struck in Jamaica at maximum intensity, one of the strongest coastal impacts ever documented in almost 200 years of data collection across the region. The confident prediction likely gave residents additional preparation time to get ready for the disaster, potentially preserving people and assets.

The Way Google’s Model Works

Google’s model operates through spotting patterns that conventional time-intensive scientific weather models may miss.

“The AI performs far faster than their physics-based cousins, and the computing power is more affordable and demanding,” said Michael Lowry, a ex forecaster.

“What this hurricane season has proven in quick time is that the recent AI weather models are on par with and, in some cases, superior than the less rapid physics-based forecasting tools we’ve traditionally leaned on,” he added.

Clarifying Machine Learning

It’s important to note, the system is an example of AI training – a technique that has been employed in research fields like weather science for years – and is not generative AI like ChatGPT.

AI training processes large datasets and pulls out patterns from them in a such a way that its model only requires minutes to generate an result, and can operate on a desktop computer – in strong contrast to the flagship models that governments have used for years that can take hours to process and need the largest supercomputers in the world.

Professional Responses and Future Advances

Nevertheless, the fact that the AI could exceed earlier gold-standard legacy models so rapidly is truly remarkable to weather scientists who have dedicated their lives trying to predict the most intense weather systems.

“I’m impressed,” said James Franklin, a retired forecaster. “The data is now large enough that it’s pretty clear this is not just beginner’s luck.”

Franklin said that while Google DeepMind is outperforming all competing systems on predicting the future path of hurricanes worldwide this year, like many AI models it occasionally gets high-end intensity predictions wrong. It had difficulty with another storm previously, as it was similarly experiencing quick strengthening to category 5 above the Caribbean.

During the next break, Franklin stated he intends to talk with Google about how it can make the DeepMind output even more helpful for forecasters by offering extra under-the-hood data they can use to evaluate exactly why it is producing its answers.

“A key concern that troubles me is that while these predictions appear highly accurate, the results of the system is kind of a black box,” said Franklin.

Wider Sector Trends

There has never been a commercial entity that has produced a high-performance weather model which grants experts a view of its techniques – unlike nearly all other models which are offered at no cost to the general audience in their full form by the governments that created and operate them.

Google is not alone in starting to use AI to solve challenging meteorological problems. The authorities also have their respective AI weather models in the development phase – which have demonstrated improved skill over previous non-AI versions.

Future developments in artificial intelligence predictions appear to involve startup companies taking swings at formerly difficult problems such as sub-seasonal outlooks and improved early alerts of severe weather and sudden deluges – and they are receiving US government funding to pursue this. A particular firm, WindBorne Systems, is even deploying its proprietary atmospheric sensors to address deficiencies in the US weather-observing network.