Sat Dec 21 02:34:30 UTC 2024: ## Google’s AI Weather Model Outperforms Traditional Methods

**Bengaluru, India (December 21, 2024)** – Google DeepMind’s new artificial intelligence (AI) model, GenCast, is generating significantly more accurate weather forecasts than existing numerical weather prediction (NWP) models, according to a recent Nature publication. GenCast leverages ensemble forecasting, utilizing an AI model trained on 40 years of reanalysis data to predict weather up to 15 days in advance with a spatial resolution of 0.25° x 0.25° and temporal resolution of 12 hours.

Unlike traditional NWP models, which provide deterministic forecasts, GenCast produces probabilistic forecasts, offering the probability of certain weather events. This probabilistic approach is particularly beneficial for predicting extreme weather events, giving more lead time for preparation. In tests against the European Centre for Medium-Range Weather Forecasts (ECMWF)’s ensemble forecasts (ENS), considered a benchmark in the field, GenCast demonstrated superior accuracy in 97.2% of 1,320 evaluated targets, and significantly better accuracy for predictions beyond 36 hours.

GenCast’s speed is another advantage. The model, which uses Google’s Tensor Processing Units (TPUs), generates a 15-day ensemble forecast in just eight minutes on a single TPU v5 unit, considerably faster than the hours required by supercomputers for NWP.

While GenCast represents a significant advancement, experts emphasize that traditional NWP models remain crucial. They provide the foundational weather data and are essential for understanding the rapidly changing global climate, which presents challenges beyond the capabilities of historical data alone. Google has affirmed its commitment to collaborating with weather agencies to integrate AI methods with traditional approaches. The code for GenCast is publicly available on GitHub. This development joins other breakthroughs in AI-powered weather forecasting, including Huawei’s Pangu-Weather and Nvidia’s FourCastNet, demonstrating the growing potential of AI in improving weather prediction accuracy and speed.

Read More