Improving rainfall and water-cycle projections through machine learning
Changes in the global water cycle due to greenhouse gas emissions can have a major impact on society and ecosystems. However, climate models struggle to accurately simulate the global water cycle, making it difficult to predict the climate-change response. Common problems include errors in the intensity of rainfall (including extreme rainfall) and the surface winds that influence evaporation and the ocean circulation. We will discuss efforts to develop new machine-learning components of climate models to help address these problems. We will focus on approaches that are physically consistent, stable and robust, and which can be applied over a range of spatial resolutions.