NEW ORLEANS – The year is coming to an end. One of the biggest topics we’ve talked about is artificial intelligence (AI) and how it’s changing our lives right before our eyes. AI is clearly the next step in our changing digital world, but something that has always been at the cutting edge of technology is weather forecasting and computer weather models. How AI and weather forecasting work together in the future will be crucial to protecting people from natural disasters in the future.

“An additional 12 or 24 hour warning of a cyclone can be hugely important,” said Dr. Ilan Price, senior research scientist at Google DeepMind. His team is a leader in using AI to revolutionize weather forecasting.

Price’s team developed GenCast, a high-resolution AI weather model that launched in 2024. It is faster and more accurate than the European ensemble model, which is considered the global gold standard.

“We’re talking about doing in about eight minutes on a computer chip slightly larger than a laptop what traditional methods would take up to hours on a supercomputer – which has tens or hundreds of thousands of processes,” Price said.

AI weather models are different from traditional weather models. Meteorologists have been using it for years because AI doesn’t use physical equations. Instead, AI weather models learn from historical weather data.

“It can learn richer, more complex and more realistic dynamics than those described by the equations that traditional models solve,” Price explained.

Traditional weather models rely on physical calculations to simulate the behavior of the atmosphere. In contrast, GenCast uses historical data to teach itself patterns. By showing the model’s past weather conditions and asking it to predict future scenarios, researchers refine its accuracy over time.

For example, a researcher shows the model the weather at a specific time in the past and asks it to make a forecast for 12 hours in the future. They show what the model should have predicted to the model itself, so that the model understands what its mistake was and can learn from it. You repeat this process over and over again, allowing the model to learn more detailed atmospheric dynamics based on real life.

However, traditional models remain indispensable. GenCast relies on fundamental data from physics-based models. For example, ERA 5 reanalysis data is used, which combines historical observations with model results.

Basically, the better the data quality, the better the AI ​​learns. Therefore, nothing is lost, and this is just another benefit for weather forecasting overall.

GenCast predicts a range of meteorological variables at all atmospheric levels up to 15 days in advance. It excels at tracking tropical cyclones, predicting extreme heat and cold, and even supporting the renewable energy sector by predicting wind power production.

GenCast is better at predicting daily weather, extreme heat and cold, and tropical cyclone tracks than the European ensemble model, considered the industry gold standard. An ensemble model provides meteorologists with a range of weather forecast scenarios and a probability of how likely that scenario is to occur. Ensemble models communicate the uncertainty in a forecast and show many different results instead of just one – a so-called deterministic model.

Google DeepMind released its deterministic AI model called GraphCast last year. There are now more options available when using an ensemble model, meaning meteorologists can make better predictions for a hurricane making landfall or just a sunny day in the park.

“This does not replace traditional models or meteorologists, but adds another powerful tool to the forecast toolbox,” Price said.

Despite advances in AI, meteorologists remain critically important. “I don’t think there will ever be no need for a human meteorologist,” said Chris Franklin, chief meteorologist for WWL Louisiana.

Computer models have evolved for decades, and yet the need for human input has not diminished, Franklin said.

“There were probably concerns in the ’70s, ’80s and ’90s as computer models got better and more advanced,” Franklin said. “I think AI is in some ways similar to improving computer models and is more just a tool to help us make better forecasts.”

Price agreed, adding: “This doesn’t replace all physics-based models and meteorology and meteorologists, but it’s hopefully a much better tool in the toolbox going forward.”

While GenCast is groundbreaking, it has limitations. It underestimates the intensity of tropical cyclones because the historical data used for training often underestimated storm strength. It also can’t yet predict cloud cover, a capability that researchers are working on adding.

Google DeepMind is not alone in this area. Other organizations are developing AI weather models, and Price hopes industry-wide collaboration will accelerate advances in meteorology.

The GenCast model and code are currently free and open to the public and their goal is to help make people and property safer in the future.

Further information about GenCast can be found HERE.

Click here to report a typo.

► Get breaking news from your neighborhood delivered straight to you by downloading the new FREE WWL-TV News app now IOS app store or Google Play.