Personalised weather forecasting through deep learning-based text generation from weather model output

The latest enhancements of numerical weather prediction models are leading to better forecasting skill, more localised predictions and longer forecasting periods.

The technological developments allow for a new, disruptive application based on artificial intelligence (AI): personalised weather forecasts.

Currently, the KNMI forecast office issues the ‘basisverwachting’, which is a general high level country-wide state of the weather for the general public. However, weather models could in principle also be used to provide custom messaging tailored to you based on your approximate location, providing more effective and useful communication of localised weather information, especially warnings. This project explores text-based generative AI (e.g., ChatGPT) to start developing meteorological text generation capabilities, which could improve the effectiveness of disseminating weather information.

The project is funded through the KNMI Meerjarig Strategisch Onderzoek (MSO) program and runs from January 2024 to December 2026.

Examples with hallucinated weather forecasts
Examples with hallucinated weather forecasts
Example of the KNMI 'basisverwachting'
Example of the KNMI 'basisverwachting'