Field-Space Autoencoder for Scalable Climate Emulators

2026-01-21

AI Summary

Kilometer-scale Earth system models are computationally expensive and data-intensive, limiting their applicability in tasks such as probabilistic risk assessment. This work proposes a spherical-compression-based field-space autoencoder framework that operates directly on raw climate model outputs, circumventing geometric distortions inherent in Euclidean grids. By integrating diffusion generative models with field-space attention mechanisms, the approach achieves physically consistent zero-shot super-resolution while preserving structural fidelity. The method effectively leverages multi-resolution information and substantially outperforms convolutional baselines in generative climate simulation, enabling the construction of scalable, high-fidelity climate emulators.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Kilometer-scale Earth system models are essential for capturing local climate change. However, these models are computationally expensive and produce petabyte-scale outputs, which limits their utility for applications such as probabilistic risk assessment. Here, we present the Field-Space Autoencoder, a scalable climate emulation framework based on a spherical compression model that overcomes these challenges. By utilizing Field-Space Attention, the model efficiently operates on native climate model output and therefore avoids geometric distortions caused by forcing spherical data onto Euclidean grids. This approach preserves physical structures significantly better than convolutional baselines. By producing a structured compressed field, it serves as a good baseline for downstream generative emulation. In addition, the model can perform zero-shot super-resolution that maps low-resolution large ensembles and scarce high-resolution data into a shared representation. We train a generative diffusion model on these compressed fields. The model can simultaneously learn internal variability from abundant low-resolution data and fine-scale physics from sparse high-resolution data. Our work bridges the gap between the high volume of low-resolution ensemble statistics and the scarcity of high-resolution physical detail.

Problem

• climate emulation
• computational scalability
• high-resolution climate modeling
• data compression
• Earth system models

Innovation

• Field-Space Autoencoder
• spherical compression
• zero-shot super-resolution
• climate emulation
• diffusion generative model