• by Laurie Sullivan  @lauriesullivan, April 1, 2024

Weather-based ad targeting using data and analytics has been around for years, but Google researchers said they have developed a way to integrate generative artificial intelligence (GAI) to advance the ability to more accurately forecast weather.

Scalable Ensemble Envelope Diffusion Sampler (SEEDS), recently published in Science Advances, is a generative AI model that can generate groups of weather forecasts that can be accessed at a small fraction of the cost of traditional physics-based forecasting models. 

It is "computationally costly" to generate large groups of data to characterize rare and extreme weather events accurately.

One caveat is that there was no mention of using the technology for ad targeting, although it is an obvious use for the advancement. 

The company's researchers have been studying weather.

Some recent innovations include MetNet-3, Google's high-resolution forecasts up to 24 hours into the future, and GraphCast, a weather model that can predict weather up to 10 days in advance. 

The research paper outlines more of forecasting accurate weather and less about using that weather for targeting ads.

It is a well-known fact that weather has been a part of targeting factors for years.

The findings and the work of MIT meteorology professor Ed Lorenz may not be as well known. He has written papers on “very-long-range weather prediction” that describe how errors in initial conditions grow exponentially when integrated in time with numerical weather-prediction models.

This has resulted in a “deterministic predictability limit that restricts the use of individual forecasts in decision making, because they do not quantify the inherent uncertainty of weather conditions,” and has become problematic when forecasting extreme weather events, such as hurricanes, heatwaves, or floods.

Since generative AI (GAI) is known to generate very detailed images and videos, it can become useful for generating ensemble forecasts that are consistent with plausible weather patterns.

The researchers point to a statement by Lorenz that suggests "The [weather forecast] maps which they produce should look like real weather maps." It does when GAI is used

SEEDS leverages generative AI to produce these forecasts, but in comparison with operational U.S. forecast system, can do so at an accelerated pace.

The results reported in the paper needed only two seeding forecasts from the operational system, which generates 31 forecasts in its current version.

“It led to a hybrid forecasting system where a few weather trajectories computed with a physics-based model are used to seed a diffusion model that can generate additional forecasts much more efficiently,” the paper explains. “This methodology provides an alternative to the current operational weather forecasting paradigm, where the computational resources saved by the statistical emulator could be allocated to increasing the resolution of the physics-based model or issuing forecasts more frequently.”