GAIA: A Foundation Model for Operational Atmospheric Dynamics

2025-05-15

arXiv.org

AI Summary

This study addresses the challenge of learning semantically rich, atmosphere-dynamics-focused representations from geostationary satellite imagery—decoupled from diurnal texture variations—to enhance downstream atmospheric modeling. We propose the first geospatial foundation model architecture that synergistically integrates masked autoencoding (MAE) with label-free self-distillation (DINO), jointly capturing local spatiotemporal details and global dynamical dependencies. The model demonstrates robust reconstruction capability under high masking ratios and achieves high-accuracy precipitation estimation with minimal labeled data: a false alarm rate of 0.088 and structural similarity index of 0.881. Our core contribution lies in deeply adapting self-supervised learning paradigms to physical atmospheric process modeling, establishing a scalable, low-label-dependency framework for global meteorological representation learning.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Remote Sensing / Geospatial AI
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

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

Abstract

We present the GAIA (Geospatial Artificial Intelligence for Atmospheres) Foundation Model, a novel model that combines masked autoencoders (MAE) and self-DIstillation with NO labels (DINO) for analyzing global atmospheric patterns in satellite imagery. By integrating these complementary self-supervised learning approaches, our model simultaneously captures both local features and global dependencies. We address two critical challenges in satellite data analysis: reconstructing missing regions and estimating precipitation patterns as our first downstream tasks. The model demonstrates superior temporal pattern capture compared to standard MAE approaches, while maintaining robust performance in downstream tasks. Our experimental results show strong gap-filling capabilities across varying mask ratios and accurate precipitation estimation with limited training data, achieving a false alarm ratio of 0.088 and structural similarity of 0.881. This work represents an advancement in self-supervised learning for atmospheric science, providing a foundation for improved weather monitoring and climate analysis. The trained model weights and accompanying code are publicly available as open-source on Hugging Face here: https://huggingface.co/bcg-usra-nasa-gaia/GAIA-v1.

Problem

• Developing hybrid self-supervised model for atmospheric representation learning
• Capturing atmospheric dynamics beyond trivial diurnal satellite patterns
• Enhancing downstream tasks like cyclone detection and river segmentation

Innovation

• Hybrid MAE-DINO model for satellite imagery
• Learns disentangled atmospheric dynamics representations
• Achieves robust reconstruction with superior gap-filling