Using AI to Accelerate Accurate Weather Forecasting

About the Session

Traditional weather forecasting is accomplished by physics-based simulation approaches, requiring tremendous computational power to conduct even individual forecast runs faster than real-time to have predictive value. These simulations are expensive, time-consuming (but urgent), and have diminishing accuracy over longer forecast times due to the nonlinear dynamics involved and the initial data coarseness and quality. However, we have vast quantities of historical data on weather patterns and events, presenting an opportunity to train AI models on weather patterns and thus transform weather forecasting to inferencing–with accuracy and greatly reduced computational costs and time. Google’s WeatherNext is one of the first AI-based forecasting applications, and has demonstrated excellent results compared to best-in-class weather codes. Join us to learn more about how AI is being used in a historically simulation-based domain, and what this might mean for other applications and domains.