The Japan Meteorological Agency is considering introducing artificial intelligence into weather forecasting to improve accuracy by using deep learning, according to a source close to the matter.

Preparations are underway, with a new team established in April to lead infrastructure and technological development to combine AI predictions based on historical weather data with the numerical forecasting models currently in use.

If implemented, the AI system would use deep learning to analyze vast amounts of data, identifying patterns in weather to generate forecasts for future temperatures and rainfall, among other predictions.

AI-based methods may prove more accurate than current forecasting approaches in certain areas, such as typhoon path prediction.

The agency is expected to announce additional measures to enhance data functionality using AI and other technologies around June.

It currently uses a numerical weather prediction model utilizing observational data and supercomputers to simulate future atmospheric conditions. Experienced forecasters then analyze the results, factoring in real-world situations to issue weather reports and disaster alerts.

AI has so far been used to fine-tune simulation outputs, and humans will still play an oversight role and issue final forecasts even if AI predictions are introduced.

In addition to weather forecasting, the agency is exploring using deep learning to process observational data from the Himawari-10 weather satellite, set to begin operation in fiscal 2029.