Transforming global water cycle observations via synergistic AI and remote sensing

Rapid shifts in terrestrial water cycle and water-related disasters due to climate change challenge the capability of remote sensing observation. Current models that rely on region-specific empirical parameters and multistage workflows fail to robustly handle complex and accelerated water cycle with high spatiotemporal resolutions. Increasing parameter complexity for fine-scale representation under various terrestrial conditions becomes prohibitive. To address these limitations, we present bidirectional encoder representations from transformers for hydrology (BERTH), an end-to-end artificial intelligence (AI) framework for global water cycle monitoring at 30-meter and daily scale, directly transforming remote sensing radiance to evapotranspiration, precipitation, soil moisture, and runoff. As a transformative paradigm for quantitative Earth science, BERTH trained on global datasets precisely captures spatiotemporal dynamics across diverse conditions, surpassing existing geophysical models in accuracy. Embodying the synergistic integration of AI and remote sensing, BERTH offers a transformative platform that supports next-generation investigations into global water resources and Earth system dynamics.