Google Builds AI System That Can Predict Extreme Weather

Google has created a new generative AI-based system that it claims can predict weather much faster and more economically than ever before – while also detecting hard-to-spot extreme weather events.

Google claims the model can generate accurate weather forecasts at scale, with the “Scalable Ensemble Envelope Diffusion Sampler” (SEEDS) being designed similarly to popular large language models (LLMs) like ChatGPT and generative AI tools such as Sora, which generates videos from text prompts.

SEEDS generates multiple weather scenarios much faster and cheaper than traditional predicting models can, and the team described its findings in a paper published in the journal Science Advances on March 29.

🤖🇺🇸 Google has concocted an AI model named SEEDS, it can whip up accurate weather forecasts efficiently and at a fraction of the cost of conventional methods. pic.twitter.com/fYNZW3nfBx

— Talk AI Today (@TalkAIToday) April 22, 2024

Weather prediction is challenging, with many variables variables that can potentially lead to catastrophic weather events, from hurricanes to heat waves. With climate change worsening, and extreme weather events becoming more frequent, the ability to accurately predict the weather could save lives by giving people enough time to prepare for the worst effects of natural disasters.

Physics-based predictions currently employed by weather services gather various measurements and provide a final prediction that averages many different modeled predictions, based on all the available variables. Rather than a single forecast, weather forecasting is based on a set of predictions per forecast cycle that provides a range of possible future scenarios.

This means that most weather predictions are accurate for more common scenarios such as light cloud cover or warm summer days, but generating enough forecast models to find the likely outcome of an extreme weather event is not possible for most services.

Current predictions are based on deterministic or probabilistic forecast models, in which random variables are introduced to the initial conditions. However, this results in a higher error rate, meaning that accurately predicting extreme weather and weather further in the future is difficult to get right.

Unforeseen errors in the initial conditions can also greatly affect the prediction result as the variables grow over time, and modeling enough forecasts to account for variables down to such intricate detail is very costly. The scientists at Google estimated that 10,000 predictions in a model are needed to forecast events that are only one percent likely to happen.

SEEDS creates prediction models from physical measurements collected by weather agencies. It studies the relationships between the potential energy unit per mass of Earth’s gravity field in the mid-troposphere and sea level pressure – two common measures used in forecasting.

Traditional methods typically produce ensembles of about 10 to 50 predictions. However, by using AI, the current version of SEEDS can extrapolate up to 31 prediction ensembles based on just one or two “seeding forecasts” used as the input data.

The Google scientists tested the AI system by modeling the 2022 European heatwave using historical weather data recorded at the time. Just seven days before the heatwave, the US operational ensemble prediction data gave no indication that such an extreme event was incoming, Google representatives wrote in a blog post on its research portal. Adding that ensembles with less than 100 predictions – which is more than typical – would also have missed it.

The researchers described the computing costs associated with carrying out calculations with SEEDS as “negligible” compared with today’s methods. Google claims its AI system also had a throughput of 256 ensembles for every three minutes of processing time in a sample Google Cloud architecture, which can be scaled easily by recruiting more accelerators.

Extreme Weather Events in Greece, and the Applicability of Google’s New AI System

Wildfires ravaged Greece in 2023, fueled by the hot and dry conditions caused by heat waves. The islands of Rhodes and Evia were particularly affected. On Rhodes, thousands of tourists were evacuated from local resorts as the wildfire burned out of control for several days.

In late August, a wildfire in Evros in northeastern Greece burned uncontrollably for weeks and was declared the largest the EU has ever faced. It destroyed homes and caused multiple evacuations of villages and the city’s hospital.

Floods followed the devasting wildfires in September. The flooding in Thessaly, Greece’s worst on record, devastated the fertile region, swept away agricultural land, roads, and railways, and killed 16 people.

It was the second major flood in three years to hit Thessaly, part of a pattern of worsening extreme weather in Europe. Commenting on the the extreme weather phenomena, Greek PM Kyriakos Mitsotakis said at the time “Greece is facing a war in a time of peace.”

Google’s new SEEDS forecasting technology may – if applied on a large scale – save people from future extreme weather events like these in Greece.