Boosting (sub) seasonal forecasts with Explainable AI

In this 5-day workshop we will explore how explainable AI can be used to boost subseasonal-to-seasonal (S2S) forecast skill focusing on drought in the Horn of Africa.

In this workshop we will unite the relatively small but rapidly growing community working in this field, consisting of climate scientists, data scientists and forecast experts. In addition we will invite selected forecast users from the Horn of Africa.

Scientific goals of the workshop are to:

• Get an overview of explainable AI (XAI) methods and possible applications. During hands-on coding sessions, we will use XAI methods to:
  • Discover and understand sources of predictability
  • Improve upon benchmark numerical models
  • Explore the limits of these methods
• Kick-off a 3-year project (AI4S2S) that aims at developing community software that improves the transparency and reproducibility of ML for S2S applications. We would like to gather input on the needs of the S2S research community and start working on a set of best practices.

We aim at having lectures and discussions in the morning, with in the afternoon practical coding sessions on drought predictions for the Horn of Africa.