Traditional Numerical Weather Prediction (NWP) has two main steps: Data Assimilation to estimate the current state of the weather, and Forecasting to roll that forward in time. Because both steps rely on complex physical models running on supercomputers, only a handful of organizations in the world have been able to build and run traditional NWP models.
In the last few years, AI models have emerged as a very promising alternative for the Forecasting step, but critically, these AI models still depend on traditional Data Assimilation to initialize them with the current state of the weather. This dependency restricts when the forecasts can be run and which observational datasets can be used to drive them, thereby limiting the otherwise transformative impact of AI weather modeling.
With this in mind, Brightband has developed an AI Data Assimilation system, AIDA, that we believe is a significant step forward for AI weather prediction. AIDA is able to map directly from Level 1 sensor data – effectively raw data – to the current atmospheric state, without reference to a background state.
AIDA’s output can serve as the input to any number of AI Forecast Models, enabling a fully AI-driven “observations-to-forecast” pipeline. And by bypassing traditional DA and using only fresh observational data, an AIDA-driven forecast can deliver the weather forecast earlier than Forecast Models initialized by traditional DA.
We are happy to share that AIDA is running operationally, and we’re excited to refine and extend this system in the coming months.
If you’re interested in developing your own observation-driven AI models, check out Brightband’s NNJA-AI data release, which we hope will foster new research at this exciting frontier.
Why AIDA? (other than standing for Artificial Intelligence Data Assimilation) Brightband CEO Julian Green played trumpet 🎺 as a kid and enjoyed belting out the triumphal march from the AIDA opera, so we could not resist this particular acronym and reference. In the opera, all the instruments in the orchestra and voices in the chorus build together to a triumphant celebration. Hopefully AIDA’s weaving together of different weather observations to estimate the global weather state is similarly successful.
