SatExt: Generative AI framework for Spatio-Spectral Satellite Unification and Beyond

Satellite imagery is indispensable for remote sensing and environmental monitoring. However, despite the presence of over a thousand Earth-observation satellites, achieving frequent and consistent global monitoring remains challenging due to the heterogeneity across satellite constellations. This work explores the potential of generative AI as a foundation for unifying such heterogeneous imagery into a standardized, high-resolution format. We introduce SatExt, a modular generative framework that enables spatio-spectral unification of multi-satellite data to facilitate frequent and consistent Earth monitoring. SatExt strategically decouples the unification process into two stages: spectral extension and spatial super-resolution. It integrates a lightweight attention-based network for spectral extension with a diffusion-based model for high-resolution spatial reconstruction. We evaluate SatExt on representative heterogeneous-satellite dataset and demonstrate promising quantitative and qualitative performance. Through this investigation, we highligh key challenges and exciting opportunities toward making satellite imaging more reliable, consistent, and unified for global Earth observation.