Learning from the high resolution data, as well as from model errors (Data Assimilation Increments), we will develop Machine Learning algorithms that are physics- and scale-aware to accelerate climate science discovery .
Our transformative methodology will rely on a range of ML algorithms (e.g., convolutional neural networks, equation-discovery), including with strict physical constraints, or with embeded physics for improved interpretability, and generalizations.
Those ML models will be used to deepen our understanding of subgrid processes, and will be implemented as subgrid parameterizations or as bias corrections, in coarser resolution ocean (NEMO and MOM6) and coupled climate models (IPSLCM, CM4, CESM).
Learn more:
Learn about Blending Physics and Machine Learning for Climate Projections in this talk with Laure Zanna, Lead PI of M²LInES
Learn about physics-guided Machine Learning in this talk by Pierre Gentine, PI at M²LInES
Learn about discovering equations from data for ocean mesoscale parametrizations here with Laure Zanna, Lead PI of M²LInES
