Geophysical processes are typically sparsely observed and the relations between observed quantities and geophysical variables are often nonlinear, so that the statistics are non-Gaussian. Data assimilation (DA), data fusion, machine learning (ML), uncertainty quantification (UQ) and predictability are important areas of active research in geoscience. This interdisciplinary session focuses on new ideas and advanced techniques across geoscience that improve the robustness and the efficiency of computational methods for fusing models and data. Our session is broad in the application areas, and numerical methods of interest include variational and ensemble DA, Markov chain Monte Carlo, ensemble-based and adjoint sensitivity, observation impact and targeting, particle filters, deep/reinforcement learning, inversion techniques, deep/statistical/machine learning from scarce data, control variable transforms/pre-conditioning, diffusion models as well as Gaussian processes. We also welcome contributions to new computational infrastructure, e.g., the Joint Effort for Data assimilation Integration (JEDI).

• A - Atmospheric Sciences
• GC - Global Environmental Change
• IN - Informatics
• OS - Ocean Sciences