Towards diffusion models for large-scale sea-ice modelling

Tobias Sebastian Finn,Charlotte Durand,Alban Farchi,Marc Bocquet,Julien Brajard

2024-07-22

Abstract:We make the first steps towards diffusion models for unconditional generation of multivariate and Arctic-wide sea-ice states. While targeting to reduce the computational costs by diffusion in latent space, latent diffusion models also offer the possibility to integrate physical knowledge into the generation process. We tailor latent diffusion models to sea-ice physics with a censored Gaussian distribution in data space to generate data that follows the physical bounds of the modelled variables. Our latent diffusion models reach similar scores as the diffusion model trained in data space, but they smooth the generated fields as caused by the latent mapping. While enforcing physical bounds cannot reduce the smoothing, it improves the representation of the marginal ice zone. Therefore, for large-scale Earth system modelling, latent diffusion models can have many advantages compared to diffusion in data space if the significant barrier of smoothing can be resolved.

Machine Learning,Atmospheric and Oceanic Physics

What problem does this paper attempt to address?

The paper aims to explore how to use Latent Diffusion Models (LDMs) to generate large-scale Arctic sea ice state data. Specifically, the research objectives include:

1. **Reducing computational cost**: By performing diffusion in the latent space rather than directly operating in the high-dimensional data space, thereby reducing the demand for computational resources.
2. **Integrating physical knowledge**: Incorporating physical constraints into the generation process to ensure that the generated data conforms to physical laws. For example, by applying truncated Gaussian distributions to the encoder and decoder to generate data that meets physical boundary conditions.
3. **Evaluating the performance differences between latent diffusion models and data space diffusion models**: The study found that the results generated by latent diffusion models are relatively smoother, but the performance in edge areas has improved. Nevertheless, the smoothing issue remains a major obstacle that needs to be addressed.
4. **Improving generation speed**: Latent diffusion models significantly outperform diffusion models in the data space in terms of generation speed, which helps save time and computational resources in practical applications.

In summary, the key issue that this paper attempts to address is: how to efficiently generate large-scale sea ice state data using latent diffusion models while ensuring physical consistency and overcoming the smoothing problems brought by latent space mapping.