DiffDA: a Diffusion Model for Weather-scale Data Assimilation
This work addresses the need for accurate data assimilation in weather forecasting and climate modeling, offering a novel method that could enable real-world applications like reanalysis datasets, though it is incremental as it builds on existing neural network backbones.
The authors tackled the problem of generating initial conditions for weather forecasting by proposing DiffDA, a diffusion model that assimilates atmospheric variables from sparse observations and predicted states, achieving the highest resolution (0.25 deg globally) for ML data assimilation models and reducing lead time loss to at most 24 hours compared to state-of-the-art methods.
The generation of initial conditions via accurate data assimilation is crucial for weather forecasting and climate modeling. We propose DiffDA as a denoising diffusion model capable of assimilating atmospheric variables using predicted states and sparse observations. Acknowledging the similarity between a weather forecast model and a denoising diffusion model dedicated to weather applications, we adapt the pretrained GraphCast neural network as the backbone of the diffusion model. Through experiments based on simulated observations from the ERA5 reanalysis dataset, our method can produce assimilated global atmospheric data consistent with observations at 0.25 deg (~30km) resolution globally. This marks the highest resolution achieved by ML data assimilation models. The experiments also show that the initial conditions assimilated from sparse observations (less than 0.96% of gridded data) and 48-hour forecast can be used for forecast models with a loss of lead time of at most 24 hours compared to initial conditions from state-of-the-art data assimilation in ERA5. This enables the application of the method to real-world applications, such as creating reanalysis datasets with autoregressive data assimilation.