Multimodal atmospheric super-resolution with deep generative models

Diffusion models are a class of generative machine learning algorithms that can be used to sample from complex distributions. They achieve this by learning a score function, i.e. the gradient of the log-probability density of the data, and reversing a noising process using the same. Once trained, these diffusion models not only generate new samples but also enable zero-shot conditioning of the generated samples on observed data. This promises a novel paradigm for data and model fusion, wherein the implicitly learned distributions of pretrained diffusion models can be updated given the availability of online data in a Bayesian formulation. In this article, we apply such a concept to the super-resolution of a high-dimensional dynamical system, given the real-time availability of low-resolution (LR) and experimentally observed sparse measurements from multimodal data. Our experiments are performed for a super-resolution task that generates the ERA5 atmospheric dataset given sparse observations from a coarse-grained representation of the same and/or from unstructured experimental observations of the Integrated Global Radiosonde Archive radiosonde dataset. We also perform a data fusion task that leverages predictions from a data-driven atmospheric emulator as well. We discover that the generative model can balance the influence of multiple dataset modalities during spatiotemporal state reconstructions. Additional analysis on how score-based sampling can be used for uncertainty estimates is also provided.