Using Diffusion Models to do Data Assimilation (2506.02249v1)

Abstract: The recent surge in ML methods for geophysical modeling has raised the question of how these same ML methods might be applied to data assimilation (DA). We focus on diffusion modeling (a form of generative artificial intelligence) and on systems that can perform the entire DA, rather than on ML-based tools that are used within an otherwise conventional DA system. We explain that there are (at least) three different types of diffusion-based DA systems and we show in detail that the three systems differ in the type of posterior distribution they target for sampling. The different posterior distributions correspond to different priors and\slash or likelihoods, which in turn results in different types of training data sets, different computational requirements and different qualities of their state estimates. We discuss the implications of these findings for the use of diffusion modeling in DA.