Day 1: Peter Dueben - Machine Learning and Earth System Modelling | ADIA Lab Symposium 2025

Value of the Information & Strength of the Argument

The talk provides valuable insights into the current state and future direction of machine learning in Earth system modelling. Dueben’s argumentation is solid, grounded in his experience at ECMWF and ongoing projects. He effectively contrasts the quiet revolution with the ML revolution, using concrete examples like the Dubai flood event to illustrate ML’s capability. He also presents a balanced view, acknowledging the limitations and risks of ML, such as data scarcity for climate and the difficulty of verifying ML predictions. The argumentation is persuasive and well-structured, though some points are speculative and based on personal projections rather than empirical evidence.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates scientific rigour through the speaker’s expertise and the use of real-world examples. However, specific sources are not cited within the talk; the only external reference is the ADIA Lab symposium link in the description. The title accurately reflects the content, and the talk is well-organized. The speaker’s credibility as a leading expert adds to the reliability, but the lack of formal citations limits the verifiability of some claims.

The title accurately reflects the content: a talk on machine learning in Earth system modelling, delivered at the ADIA Lab Symposium 2025.

8/10

The speaker is a leading expert from ECMWF, providing a credible overview of the field. The talk is based on his professional experience and ongoing projects, but it is not a peer-reviewed study. The content is well-structured and technically accurate, though some claims are forward-looking and not yet fully validated.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 01:00Introduction: Peter Dueben begins his talk on machine learning and Earth system modelling.
• 03:00Shows a video of high-resolution Earth system model output, highlighting the success of physical models.
• 05:00Discusses the improvement in weather forecast skill over decades, the 'quiet revolution'.
• 08:00Introduces the machine learning revolution and its impact on forecast skill, with a sudden jump in scores.
• 10:00Explains the use of machine learning in operational weather forecasting, including the AIFS model and hybrid approaches like nudging.
• 12:00Discusses the need for foundation models and introduces the WeatherGenerator project.
• 15:00Addresses the question of data availability for different Earth system components, highlighting limitations for climate.
• 18:00Emphasizes the continued importance of physical models for generating training data and ensuring physical realism.
• 20:00Outlines the future of Earth system modelling, with a shift towards ML and the need for critical evaluation of ML outputs.
• 22:00Concludes with a vision of future user interactions with weather and climate information through ML tools.

• ADIA Lab Symposium — Event page for the symposium where this talk was presented.

• WeatherGenerator — Project website for the foundation model initiative discussed in the talk.
• ECMWF AIFS — ECMWF's operational machine learning model for weather forecasting.