Reconstructing historical climate fields with deep learning

2023-11-30

Science Advances

AI Summary

Historical climate fields suffer from extensive spatiotemporal missingness due to sparse early observations, severely hindering climate analysis and modeling. To address this, we propose a deep learning framework based on Fourier-domain convolution, trained exclusively on climate model output to reconstruct irregular, large-scale missing regions with high fidelity. This work introduces Fourier convolutional neural networks to cross-spatiotemporal-scale climate field inpainting for the first time, enabling generalization to unseen mask patterns and super-resolution reconstruction beyond training resolution. Our method accurately reproduces key climate phenomena—including El Niño and La Niña—and consistently outperforms kriging interpolation and state-of-the-art machine learning baselines across multivariate reconstruction tasks. It establishes a scalable, robust paradigm for reconstructing sparse historical climate data, offering improved physical consistency and generalizability compared to conventional statistical or purely data-driven approaches.

Abstract

Historical records of climate fields are often sparse because of missing measurements, especially before the introduction of large-scale satellite missions. Several statistical and model-based methods have been introduced to fill gaps and reconstruct historical records. Here, we use a recently introduced deep learning approach based on Fourier convolutions, trained on numerical climate model output, to reconstruct historical climate fields. Using this approach, we are able to realistically reconstruct large and irregular areas of missing data and to reproduce known historical events, such as strong El Niño or La Niña events, with very little given information. Our method outperforms the widely used statistical kriging method, as well as other recent machine learning approaches. The model generalizes to higher resolutions than the ones it was trained on and can be used on a variety of climate fields. Moreover, it allows inpainting of masks never seen before during the model training.

Problem

• Reconstruct sparse historical climate data
• Improve accuracy over traditional statistical methods
• Handle large irregular gaps in climate fields

Innovation

• Deep learning with Fourier convolutions
• Trained on numerical climate models
• Inpaints unseen masks, generalizes resolutions