Geophysical processes are typically sparsely observed, and their dynamics are nonlinear, so that the statistics are non-Gaussian. Successful and efficient data assimilation (DA), uncertainty quantification (UQ) and prediction require that these characteristics are directly addressed. This session focuses on ideas and advanced techniques for DA, UQ and predictability, for example variational methods, (ensemble) Kalman filtering, adjoint and ensemble sensitivities, hybrid techniques, NDEnVAR, representer based approaches, Monte Carlo sampling, particle filters, and linear approximations to non-Gaussian/nonlinear applications. We welcome contributions from all of the geosciences and the mathematical and statistical communities. Our main goal is to share ideas that can be applied across Earth science and the larger scientific community.