Google’s latest GenAI tool, SEEDS, enhances weather forecasting efficiency, cutting costs and accelerating prediction times.

Google has developed a groundbreaking GenAI model, the Scalable Ensemble Envelope Diffusion Sampler (SEEDS), significantly outpaces traditional weather forecasting methods. Unlike the common physics-based approaches that rely heavily on collecting and averaging various measurements, SEEDS generates multiple weather scenarios quickly and economically.

SEEDS stands apart by employing a strategy akin to that used in LLMs. It creates numerous ensembles or sets of potential weather outcomes, enabling it to handle the unpredictable nature of extreme weather conditions more effectively. This capacity is increasingly vital as climate change leads to more frequent and severe weather events.

Traditionally, weather predictions are formed from a vast array of data leading to a single forecast based on numerous possible outcomes. However, this method struggles with accuracy for rare or extreme conditions due to the sheer computational power required. SEEDS, by contrast, leverages AI to predict severe weather scenarios by analyzing relationships between atmospheric conditions and sea level pressures, making it adept at identifying less likely events.

The efficiency of SEEDS is highlighted by its ability to produce up to 31 prediction ensembles from minimal input data, a significant improvement over the 10 to 50 typically generated by traditional methods. Google’s tests, including simulations of the 2022 European heatwave, have demonstrated SEEDS’s superior predictive capabilities, particularly for events that traditional models might miss.